In this episode, Chris, Rod, and Max engage with Kurt Muehmel from Dataiku to explore the company’s approach to AI and B2B sales. They discuss Dataiku’s Universal AI Platform, its strategies for building trust with enterprise customers, and the integration of generative AI technologies.

The conversation also touches on current use cases and the challenges organizations face in adopting AI solutions, particularly regarding data quality and access. In this conversation, Kurt Muehmel discusses the challenges and strategies involved in software adoption within large organizations, emphasizing the importance of empathy and understanding user needs.

He shares insights from his journey from environmental consulting to the tech industry, highlighting the role of critical thinking in the age of AI.

Kurt also explores the impact of small teams on startups and the evolving landscape of B2B technology, urging a respectful approach to organizational requirements for successful integration.

Chapters

• 00:00 Introduction to AI Trends and Dataiku

• 01:46 Understanding Dataiku’s Universal AI Platform

• 07:05 Building Trust with Enterprise Customers

• 14:08 Integrating Generative AI into Dataiku

• 20:44 Current Use Cases and Adoption Challenges

• 28:07 Navigating Change Management in Software Adoption

• 30:52 Lessons from Rolling Out Software in Large Organizations

• 33:26 Kurt’s Journey: From Philosophy to Tech

• 39:27 The Role of Critical Thinking in the Age of AI

• 46:07 The Impact of Tiny Teams on Startups and Innovation

• 51:23 Building for B2B: Insights and Expectations

Takeaways

1. Successful enterprise AI adoption requires a universal platform approach that connects diverse data technologies and team capabilities: Kurt explains that large organizations face unique challenges in AI adoption due to their complex data environments, security requirements, and diverse workforce skills. Dataiku’s ‘Universal AI Platform’ addresses these challenges by providing connectivity to various data sources and AI technologies (from traditional ML to modern LLMs), governance frameworks for security and compliance, and multiple interfaces (full code, low code, and no code) to enable different team members to collaborate. This approach recognizes that the most successful companies will be those that can combine different technologies and skill sets within a single governance framework, rather than siloing capabilities across separate tools and teams.

2. The evolution of enterprise AI is moving from isolated use cases to orchestrated agent systems that require new governance frameworks: The podcast highlights how enterprise AI adoption has evolved from basic analytics to machine learning to today’s generative AI applications. Kurt describes how organizations initially focused on creating enterprise chatbots and RAG systems, but are now moving toward AI agents that can autonomously perform tasks. This shift introduces new challenges around what Kurt calls ‘agent orchestration’ or ‘agent ops’ - the management of multiple AI agents working together as a workforce. This requires sophisticated monitoring, controls, and governance to ensure agents operate appropriately, comply with security requirements, and remain cost-effective. The future belongs to organizations that can successfully manage this transition from single-agent use cases to orchestrated multi-agent systems that deliver differentiated value.

3. Technology adoption is fundamentally a human challenge that requires empathy rather than technical superiority: One of Kurt’s most surprising lessons from helping large organizations adopt new technology was that ‘some people like really bad software.’ He explains that regardless of how objectively superior a new technology might be, users who have relied on familiar tools for years often resist change. Kurt emphasizes that successful technology adoption requires profound empathy - acknowledging the comfort users have with existing systems while helping them transition to new approaches in ways that minimize disruption. This insight extends beyond enterprise software to broader AI adoption, suggesting that the most successful AI implementations will be those that consider the human elements of change management alongside technical capabilities. As AI continues to transform how we work, this human-centered approach to technology adoption becomes increasingly important.

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Welcome

Chris (00:01.09) Welcome to another episode of the Chris Rod Max show where every week we speak about AI news, AI trends, and connect with amazing founders and innovators in the AI space to bring you the latest and greatest in innovation. I’m really excited to be joined by my co-hosts Rod and Max, and our special guest today, Kurt Muehmel. Kurt, we’re delighted to have you on the show!

Kurt Muehmel (00:32.323) It’s my pleasure to be here. Thanks Chris, Rod, and Max for having me.

Chris (00:37.998) Amazing. Today we want to explore your experience with go-to-market strategies and B2B sales. You currently work for Dataiku, and we’d love to understand what your company does exactly. As a later-stage company, you’ve likely gained significant experience with go-to-market approaches and B2B customers.

We also want to learn about your personal journey and career path that led you to where you are today. And of course, we’ll get your perspective on some recent AI trends and developments. Does that sound good?

Kurt Muehmel (01:24.793) That sounds fantastic. I’m honored to think that my perspective might be interesting to your audience. Thank you very much for the opportunity.

Dataiku’s Business Model and Market Position

Chris (01:34.222) Great, so why don’t you start by telling us more about what Dataiku actually does and how you differ from similar players in the market?

Kurt Muehmel (01:46.947) Absolutely. So Dataiku — and regarding pronunciation, there are many schools of thought, so we never correct anyone. At Dataiku, we call ourselves the Universal AI Platform. We offer our customers the ability to control all of their different AI talents, processes, and technologies so they can create all the analytics, models, and agents their company needs.

To explain further, it’s important to understand our target customers. We primarily focus on the Global 2000 companies — the world’s largest 2000 companies. These are big organizations with significant amounts of data but also substantial constraints in terms of security, privacy, and regulations. Our customers span various industries: banking, insurance, manufacturing, retail, and more. Many are companies you interact with daily as a consumer, while others operate more in the B2B space.

These organizations need a way to connect to all their different data sources and the various technologies they use to work with that data — everything from data storage and compute to AI services from their preferred providers. They also need ways for their diverse team members to collaborate on that data, enabling both their most advanced users (data engineers and data scientists) as well as business experts who may not have “data” in their title.

Dataiku provides that platform with connectivity, governance, and multiple interfaces — full code, low code, and no code — so teams can work together to build solutions with their data.

As for how we’re different from other market players — well, there are many players, and Dataiku’s positioning is relatively broad. That’s what we mean by calling ourselves the “universal AI platform.”

Many of Dataiku’s partners like Snowflake, Databricks, and major cloud providers have functionality that overlaps with Dataiku. However, Dataiku doesn’t provide the storage and compute layer — we definitely need them. What we add on top is additional functionality to enable non-expert users. We handle much of the lower-level plumbing so organizations can focus on developing new use cases rather than getting different services to work together.

We really offer customers the opportunity to keep their options open. One challenge some customers face is that starting with a single cloud provider can create a closed ecosystem they’re stuck in. One of Dataiku’s core philosophies is that new technologies will emerge, and organizations will want to leverage these innovations. Dataiku provides that connectivity and optionality so they can adopt new technologies when appropriate.

We’ve seen this evolution during Dataiku’s nearly 13-year history. When we started, the main data paradigm was Hadoop — the big challenge was helping people use Hadoop without writing MapReduce code themselves. This evolved briefly to cloud-based Hadoop, then to object storage and distributed cloud computing. Now we have data lake, data warehouse, and data lakehouse providers, plus technologies like Kubernetes for distributing compute. And of course, we’ll discuss the emergence of AI shortly.

Ultimately, what makes Dataiku different is that optionality, the ability to enable users of all types, and our core philosophy of governance built in from the ground up.

Building Trust with Enterprise Customers

Chris (07:05.01) Thank you for that overview. So essentially, Dataiku is a day-to-day platform that serves as an interface for multiple stakeholders and users, but more importantly, it’s a way to integrate new technologies for a company. And you’re targeting the biggest players in the market. My next question is: how do you build trust with these organizations? Given that many large companies already work with tech giants like Microsoft or Oracle, how do you convince them to choose your solution?

Kurt Muehmel (07:45.945) That’s been a question for Dataiku from the very beginning. I had the pleasure of joining Dataiku early on, initially as a seller. These were literally the first discussions I had my first week — customers asking me those exact questions. Ten years later, that remains the same challenge whenever our sales team approaches a new customer.

The saying has evolved over the years, but people still say “Nobody ever gets fired for buying…” — it used to be IBM, then Oracle, now Microsoft. Those are excellent, safe choices. So what we always do is establish proof in the sales cycle — proof that our technology claims are valid and backed by the product.

Culturally as a company, from a product and marketing perspective, we’re actually very conservative. We focus on talking concretely about what our product does today. We’re not selling a future or a dream — we’re really pragmatic, perhaps sometimes to a fault. But that helps establish credibility.

Other forms of proof come from existing customers. Every software vendor has customer case studies with impressive logos and ROI numbers, so that’s not differentiating. But what we do, when necessary, is introduce prospective customers to existing customers, ideally in the same industry or similar role, so they can hear the story directly from their peers. We can do this in our enterprise sales motion, and thankfully, our existing customers are often motivated to help, which is tremendously valuable. There’s a legitimacy that comes from that peer validation.

A third important source of trust is third-party validation, particularly from analyst firms like Gartner. Dataiku is proud to be a leader in Gartner’s Magic Quadrant for several consecutive years. We invest significant time ensuring analyst firms thoroughly understand Dataiku so their analyses accurately reflect who we are.

Of course, if we go back ten years to when we were just starting, we didn’t have much of that. So what did it take to go from zero to one? I’ll be honest — there was a certain degree of luck. For Dataiku and me specifically, that luck came in finding someone at a large pharmaceutical company in the United States. This person saw part of his role as seeking out promising startups and allocating some budget to test their software, viewing it as a way to make the startup ecosystem more dynamic.

We connected after he discovered some videos Dataiku had published. That was the beginning of an opportunity for him to evaluate our technology with his teams, establish trust, and ultimately spend some budget on Dataiku. That first step was crucial in building credibility with other customers, as we could now say we were trusted by a recognizable company in a highly regulated market.

I think it’s essential for early-stage companies to find those champions. They exist, but there’s an element of luck involved, along with hard work. And obviously, you must respect their trust — trust is hard earned but easily lost. This philosophy extends to how we run our technical support team; we’re maniacal about providing exceptional support. That’s another way to reinforce trust and prevent it from gradually eroding until a supporter no longer advocates for you.

Finding the First Customer

Chris (13:19.758) I always feel like with enterprise business models, it’s a chicken and egg problem. You need references from other companies and experts, but you’ve got to get your foot in the door first. It sounds like you were fortunate or very sophisticated in your approach to that first use case that made things easier. Really interesting.

So focusing on today and AI technologies — agents being a significant development — tell us what you’re doing in this field. How are you integrating AI technology into your connectivity layer?

Integrating AI and LLMs at Dataiku

Kurt Muehmel (14:13.625) If we’re talking about generative AI and particularly large language models, Dataiku actually had our first integration with GPT-3 about a year before ChatGPT was released. This wasn’t yet on most people’s radar, but as with many technologies, there was some demand from customers and curiosity on our side.

We wanted to explore what we could do with these new language models alongside the traditional machine learning models that were the primary focus on our platform. So we created a plugin for natural language text generation connecting to GPT-3, which saw moderate use. But the interest and attention fundamentally changed when ChatGPT arrived.

Our approach then was first and foremost to talk with our customers. Many came to us after their winter holidays because their boards of directors were asking, “What’s our AI strategy?” — the board members had been playing with ChatGPT between Christmas and New Year’s.

Customers asked us to help them understand this technology and its potential. We spent time discussing and exploring possible applications with them. It’s been an evolution since then, but some elements have remained stable.

One key element is broad connectivity, which came together in what we now call LLM Mesh — our connectivity layer to all the different LLMs an organization might want to use. This includes all major providers like OpenAI and Anthropic, the cloud platforms, and self-hosted options for organizations that want to run open-source or open-weights models on their own hardware.

Dataiku supports all these approaches because we see valid use cases for each. We believe the most successful companies will combine different LLMs from different providers for different applications, and having that freedom of choice will be crucial.

So Dataiku’s approach provides controlled connectivity where you can manage permissions and access to different technologies across the organization through LLM Mesh. This includes our guardrails services for cost control, safety, and content moderation — features you’ll need regardless of the use case, so you don’t want to redevelop them for each application.

Beyond the connectivity layer, we provide tools for building with generative AI, including a prompt engineering environment called Prompt Studios. More recently, we’ve focused on agent development capabilities.

Agents represent a new modality for building with generative AI. Initially, the primary format was chatbots — organizations wanted private versions of ChatGPT. Then they wanted document search with RAG capabilities, which Dataiku addressed with our Dataiku Answers product and built-in RAG functionality.

Now we see the emergence of agents, which I believe will be the primary use case for LLMs in enterprises going forward because it’s so flexible and broad. Dataiku provides the ability to create and provide tools to agents, with interfaces for building agents through both visual no-code and coding approaches — because our customers need both options for different users.

Moving forward, the focus is shifting from building individual agents to managing a workforce of agents — entering the field of agent orchestration or “agent ops.” This includes designing how agents interact, creating multi-agent systems, implementing monitoring and controls, and establishing governance to ensure these agent workforces operate appropriately within IT and security requirements while optimizing for budget. One risk with agents is that if they run autonomously, they can quickly become expensive.

These are the considerations enterprises face when building and using agents. It’s relatively easy for anyone to build a single agent with off-the-shelf tools or license an agent from a provider if you have the budget. The real challenge is building agents at scale to create something truly differentiated and managing them effectively as the technology matures.

Evolution of Data Science and AI Use Cases

Rod (20:44.834) You mentioned that Dataiku has been around for a while — you brought up Big Data, which is a term I haven’t heard in quite some time. Reflecting on that, there have been multiple adoption waves: Big Data, Data Science, Deep Learning, and so on. Since Dataiku has a strong foothold in the enterprise market, I’m curious: looking at it today, what are the main use cases if we break it down as a percentage? What are enterprises doing with Dataiku, and how would you allocate percentages to different disciplines?

Kurt Muehmel (21:19.383) That’s an excellent question. What we’re seeing right now is a widening spectrum of customer use cases. Overall, companies use Dataiku for several major categories:

First, there’s modern or advanced analytics — this is extensive data work and automation, building new analytics, combining previously separate data sources, and developing analytics for teams across the organization, often feeding into business intelligence platforms. There’s a lot of analytics engineering happening in Dataiku.

Second, there’s machine learning and MLOps work, where data science teams develop and scale numerous machine learning models. This includes predictive maintenance models for industrial customers, pricing models for insurance companies, and everything in between.

Third, there’s generative AI use. If we look at volume or absolute numbers, you’d probably see a predominance of analytics use. Not because it’s necessarily the most important or valuable, but because that’s much of the work organizations need done today — what people spend most of their time doing.

Then you have a smaller but significant percentage focusing on machine learning and MLOps — extremely valuable work, but there are relatively fewer data scientists than analysts, which tilts the balance.

And now we have this massive uptake in generative AI, which is also changing how the other types of work are performed. For example, Dataiku has built-in AI-powered assistance for analytics and machine learning work. We’ve recently introduced functionality called “Stories” that uses generative AI to automatically create slide decks based on available data.

I think we’ll see the lines between these fields begin to blur. We’ll think less about particular techniques (analytics vs. generative AI) and more about outcomes: building a new metric, dashboard, or agent. And that agent might be building new dashboards — a valid use case for an analytics-oriented agent.

The diversity of use cases somewhat tracks the profile of users, with more analyst-type than data scientist-type personnel. But that doesn’t necessarily reflect where the value comes from.

Ultimately, Dataiku’s overall value is being a single platform for all these capabilities. Our customers appreciate not having to send teams to different tools depending on the task — they have one space to administer all this “data work.”

Data Quality Challenges in Large Organizations

Chris (25:23.022) Kurt, as a follow-up: big companies don’t necessarily struggle with talent to build models, but one significant challenge is data quality, with data being scattered and incomplete. What’s your take on this? How do you help with this issue? What other challenges do large organizations face when adopting a solution like yours?

Kurt Muehmel (25:54.031) You’re absolutely right that the main barrier for machine learning work, and especially now for generative AI work, is data access and quality. That’s why one of Dataiku’s core strengths is our data preparation and engineering capabilities to feed clean data into machine learning models or generative AI systems and agents.

This differentiates Dataiku from more narrowly focused solutions that might excel at ML or agent building but start with the assumption, “bring us a clean dataset.” Much of the work is actually getting to that clean dataset. Having a platform like Dataiku that connects from raw data through preparation and engineering, including a data catalog to identify appropriate datasets for different applications — that addresses a huge part of the challenge.

Regarding adoption, I hope what’s coming across is that Dataiku is positioned differently and more broadly than other platforms. Most organizations have provided different tools to different teams based on their skills — separate solutions for analytics, machine learning, and now exploring options for generative AI.

Dataiku offers a single platform for all these functions, which brings tremendous value. It’s often easy to convince someone in the right position of this value — they see the benefits of a unified environment where there’s crossover between domains, allowing for greater efficiency and easier transition between types of work, plus centralized governance.

But when you reach teams who have been narrowly focused on specific work using particular tools, there’s always a change management challenge. From an individual’s perspective, many are satisfied with their current tools, and even moving to something better still imposes a cost of change. If you rearrange the buttons someone uses for their daily work, that’s disruptive unless they fully understand both the organizational benefits and the personal advantages.

That adoption challenge means we must work closely with teams implementing Dataiku, understanding what users like about their current processes, what frustrates them, and designing a rollout strategy that will make them not just accept but enthusiastically embrace the new platform.

Changing how a large organization handles critical aspects of its business is always an uphill battle, but it’s necessary. I think we’ve gotten increasingly skilled at helping customers achieve the adoption and success they need to become happy Dataiku users.

Lessons from Enterprise Software Implementation

Max (30:24.022) Just to follow up on that, what’s the most interesting lesson you’ve learned from rolling out to some of these largest organizations? Our podcast listeners might be interested in learning about your experiences, given your extensive background. Could you share some of the most interesting or surprising lessons you’ve learned?

Kurt Muehmel (30:52.197) What surprised me most was discovering that some people genuinely like really bad software. There’s software out there that is objectively not great — outdated, lacking valuable capabilities for end users. However, when it’s what you’ve used in your job for possibly 20 years, change becomes extremely difficult.

I think the key is not approaching users with an attitude of superiority — “Here I am, your savior with modern software that will make your life better!” Instead, you need profound empathy: “I understand you’ve been using this software for 20 years and feel comfortable with it. I know your organization has decided to change, which is challenging for you, but let me help you transition as comfortably and seamlessly as possible. And I promise that on the other side, you’ll appreciate the new system.”

You must start from a position of empathy — change is difficult for everyone. Especially when it affects the software you use for six, seven, or eight hours every day. The most important thing is not to assume you have all the right answers. Always enter every interaction looking for what you can learn from that person and how you can help them become both more comfortable and ultimately more successful.

Kurt’s Career Journey

Chris (33:09.344) Empathy is definitely key. Speaking of change, we’re also curious about your personal journey and how you got to where you are today. There were probably many changes given your background. Could you share that story with us?

Kurt Muehmel (33:26.285) Sure. I was born in the US but moved to France shortly after college graduation because I had met the person who’s now my wife during college, and she’s French. I held various jobs initially, including teaching English — something every native English speaker does in a non-English speaking country. After several positions, I wanted to do something closer to my studies.

I had studied two subjects in college: philosophy and environmental sciences. Wanting to return to the environmental field, I found my way into environmental consulting and did that for about five years. But like many consultants, I eventually grew tired of that business model.

I had always been attracted to technology. These were the heyday years when Facebook and Google were exciting companies — from about 2006 to 2014, during the rise of these major tech players. The shine has diminished somewhat now, but I was enthusiastic about that world and wanted to participate in it somehow.

I was particularly drawn to the startup model — a small group of people banding together against the odds to see if they could succeed. Even if they failed, that would be a learning experience.

So I attended a Paris Startup Job Fair organized by an English-language blog about the French startup scene. I distributed my resume to various companies, and one of them was Dataiku.

Why would they hire someone with zero relevant experience? They needed someone who spoke English without a strong French accent because they had begun receiving inquiries from outside France, particularly from the United States. I think there was a mutual willingness to take a chance — me betting on Dataiku and Dataiku betting on me — that has worked out very well, certainly for me and presumably for Dataiku since they’ve kept me all this time.

Over the past decade at Dataiku, I’ve had the opportunity to perform various roles, which has been fantastic. It’s been quite unexpected — when I started, I assumed the company would go out of business within a year because that’s what most startups do, and I would have needed to be prepared for that.

Coming into the space, I didn’t have much ability to assess the product-market fit. I wasn’t a data scientist or from a data background, so I couldn’t judge whether it would be successful. I just really liked the people I was talking to — they seemed smart and hardworking but also fundamentally humane. It was an opportunity to take a leap that I thought would be low-stakes and probably short-term, after which I’d return to get a “real job.”

But indeed, this has become much more of a real job than I ever anticipated.

Cultural and Language Background

Chris (37:46.286) Interesting! And speaking of accents, when preparing for this session, I noticed your German-sounding name combined with your time in France and London. I wondered what kind of accent we’d hear, but you sound very American. You didn’t pick up any British influences.

Kurt Muehmel (38:05.88) My accent does sometimes change depending on my audience — I try to be a bit of a chameleon and match what I’m hearing. But yes, the name is very German. My paternal grandparents were German immigrants to the United States. Since I don’t really speak German, it often causes confusion when I do business in Germany. People see my name, immediately start speaking German to me, and I have to apologize that the language wasn’t genetically transmitted!

Philosophy, AI, and Critical Thinking

Chris (38:42.03) Not yet, not yet! It’s also interesting that you studied philosophy, which seems rare these days. With Rod and Max, we’ve discussed ethical questions around AI safety and disruption. I’m curious about your thoughts on AI and AGI — are people becoming lazier because they don’t have to think as much? What’s your position on critical thinking in the age of AI?

Kurt Muehmel (39:27.139) I decided to study philosophy because it was the hardest course I took in my freshman year. I enrolled in an Intro to Philosophy class because it sounded interesting, and it proved to be my most challenging course that year. I thought, “This is good — this is a worthwhile challenge.” Learning to read complex texts and write with precision and clarity seemed valuable for whatever I might do later. I think that was a good decision by my 18-19-year-old self.

The ability to read, think critically, and write remains essential. For me, writing is how I think — when I need to work through a problem thoroughly, I write it down, often by hand. My brain works best through the physical act of writing, seeing ideas on paper, sometimes sketching concepts, and structuring words into sentences, paragraphs, and essays.

When I use generative AI in my work and personal life, I’m a heavy user but not for those core thinking tasks. I might use it as a sparring partner — “What do you think of this? Are there other approaches?” — but I believe that kernel of thinking requires personal engagement. There are valid concerns that we could lose something in terms of critical thinking if everything is first summarized by an LLM and we’re just writing bullet points for it to expand.

At the same time, it’s wonderful to have tools that help understand complex philosophical concepts. In my current use, when listening to a philosophy podcast, I might ask an AI model to explain ideas and have a back-and-forth: “Is this related to that? Did these thinkers work together? What if I think about it this way?” That can be extremely productive.

So I’m not solely positive about the technology, but I see tremendous benefits as it applies to critical thinking — as long as people use it as augmentation rather than replacement for their thinking.

How this plays out is one of the main challenges facing society: how do we adopt this technology to strengthen not just our businesses but our education and broader social systems? We’ve been discussing enterprise applications, which are certainly important — enterprises provide structure and are where many people work and interact. But especially as we approach AGI, we face fundamental questions about how we interact with each other and what we expect from employment — will we continue working 40-hour weeks, or something different?

These are profound questions that this broadly applicable, general-purpose technology will force us to address as a society. What concerns me is that societal-level discussions today don’t seem to be at their most sophisticated across different governments and populations. We’re entering a volatile period regarding the roles of nation-states and citizens and their interaction with private enterprise — all increasingly mediated and influenced by ever-more-powerful AI technologies.

It’s a fascinating time to be alive! I believe this will be one of the most transformative decades in history, especially because changes are happening so rapidly. Previous industrial and political revolutions often built up and unfolded over decades or centuries, whereas this wave is breaking over us in just a few years. There’s legitimate concern, but also a need for serious thinking from every corner of society — government, private enterprise, and even religion.

AI Startups and the Future of Work

Chris (45:51.822) Honestly, we could do an entire separate philosophy podcast on these fundamental questions. But in the interest of time, I’d like to get your opinion on a recent New York Times article about startups changing Silicon Valley. The piece focuses on the concept of “tiny teams” — companies like Gamma (AI for presentations), Anthropic, or Eleven Labs that achieve remarkable annual revenue with extremely small teams, where previously you might have needed 200-400 people, especially in operations, just to engage with enterprises and convince them to adopt your platform. What’s your take on this trend, and how do you think it will shape the startup and VC landscape, given your experience?

Kurt Muehmel (47:07.237) I saw that article, and honestly, it makes me jealous that this technology wasn’t available when we were a 20-person startup! We discussed the challenges of adopting technology like Dataiku in large enterprises — large organizations move slowly with many pieces to change. Flip that perspective to a 10-40 person company: they have virtually no legacy processes or employees to retrain. Everything can pivot quickly. The opportunity for adopting transformative technology is massive within startups.

These companies offer a glimpse of the future — of new economic models. When I was functioning as both BDR and account executive at Dataiku, I wrote little macros to scrape conference attendee lists, look people up on LinkedIn, and automate messages. Today, I could do all that with an agent — faster, more effectively, and more repeatably.

These AI-first startups are building all their processes with the fundamental question: “How can I automate this? How can I use AI for this?” They assume from the outset that tasks should not be human work, and the question becomes: what must be done by humans? They assume everything should be AI, with human tasks being the exception rather than the other way around.

This previews what future enterprises will look like in many fields. There will be exceptions — human services like nursing, haircuts, dentistry will continue requiring a human touch, literally. But for work primarily done behind computers, any new company should assume everything should be AI-powered, with humans overseeing, managing, and developing those AI capabilities.

This will lead to fundamental changes in company valuations. Like many powerful technologies, this will widen the gap between the successful and less successful, both for companies and individuals. It’s a giant lever allowing you to do more with your capital. Things will move faster, and the path to success may narrow for certain companies as competition increases with more people accessing these tools.

Hopefully, this might widen participation in the startup economy beyond people in a few global hubs or those with advanced computer science degrees — extending to smart, motivated people who might live elsewhere or lack formal training. It will be extremely competitive, but with the potential for extraordinarily high valuations and massive revenue from modest starting points, leading to multiples far exceeding anything we’ve seen before.

Closing Thoughts

Chris (51:01.304) Definitely, and we’ll likely see many more startups entering the market, which is good for competition and innovation. This has been fascinating. As a final question: do you have any parting thoughts for the audience about building for B2B, perhaps connecting to philosophy and AI, before we wrap up?

Kurt Muehmel (51:21.573) For building in B2B, it’s crucial to approach the market with realistic expectations about how large organizations operate. You’ll get frustrated and fail if you assume these organizations can move quickly or take shortcuts with security. View those security questionnaires not as burdens but as critical pathways to success, which means engineering your product to anticipate these requirements. That’s key to moving beyond flashy proof-of-concept demonstrations into becoming mission-critical software. You need to respect these requirements for long-term success in this market.

At the same time, I find B2B work profoundly interesting because, returning to philosophy, it gives me the opportunity to work with diverse companies across many sectors. At Dataiku, positioned across industries, we collaborate with organizations on their most challenging problems. You see where theory meets practice — how private enterprise uses technology to achieve business outcomes in a geopolitical context while expecting human employees to perform specific functions. It’s a convergence of major themes of our time that’s truly fascinating to witness and participate in.

Chris (53:34.606) I’ve learned a lot today. You emphasized the importance of empathy, humility, and taking others seriously. And you shared your passion for understanding business problems and what’s top of mind for companies.

I really enjoyed this session. Thank you so much for joining us. We’ll link your contact information in the show notes. And to our audience, thank you for listening! If you enjoyed today’s episode, please subscribe, like, and sign up for our newsletter. We’ll see you next week.

Your Hosts

Christine Wang: https://www.linkedin.com/in/christinewang0/ 

Rod Rivera: https://www.linkedin.com/in/rodriveracom/ 

Maxson J.Y. Tee: https://www.linkedin.com/in/maxsontjy/ 

Social: X: https://x.com/chrisrodmax 

Instagram: https://instagram.com/chrisrodmax 

LinkedIn: https://linkedin.com/company/chrisrodmax 

Subscribe: https://chrisrodmax.com Tags: #chrisrodmax #ai #technews

Andrej Žukov is an expert in natural language processing and voice AI. Rod Rivera and Max Tee discuss with him the evolution of NLP, the current trends in voice AI, the challenges of adoption, and how businesses can implement voice AI effectively.

The conversation also delves into the impact of AI on pricing strategies, highlighting the importance of flexibility in billing and the changing landscape of consumer pricing.

We explore the evolving landscape of AI pricing models, the future of outcome-based pricing, Stripe’s strategic positioning in the AI billing space, and the transformative role of AI in market research and consulting.

We discuss how flexible billing systems are essential for adapting to new business models and how AI can enhance research capabilities while also reshaping traditional consulting roles.

Chapters

• 00:00 Introduction to AI and Voice Technology

• 01:10 The Evolution of Natural Language Processing

• 06:20 Current Trends in Voice AI

• 11:22 Challenges in Voice AI Adoption

• 19:22 Implementing Voice AI in Business

• 25:19 AI and Pricing Strategies

• 36:19 Evolving Pricing Models in Tech

• 41:05 The Future of Outcome-Based Pricing

• 45:21 Stripe’s Position in the AI Landscape

• 52:04 The Role of AI in Market Research and Consulting

Takeaways

Voice AI is approaching a breakthrough moment where technology and user experience finally converge: The combination of large language models handling idiosyncratic speech, effective tool-calling capabilities, and dramatic improvements in speech-to-text conversion are creating systems that people might actually enjoy using. While businesses are already adopting voice AI for low-stakes applications like debt collection, bookings, and basic customer service, high-stakes interactions still require human touch. The final barrier to widespread consumer adoption is believability – voice AI systems need to adapt naturally to conversational context rather than using artificial, one-size-fits-all voices and tones.

AI is fundamentally changing pricing strategies, but billing infrastructure often constrains innovation: As businesses explore outcome-based pricing for AI services, they’re constrained by what their billing systems can actually handle. For consumer AI, the market has oddly settled on $20/month as a standard price point, while business applications naturally gravitate toward usage-based models with committed bands and volume discounts. As billing systems evolve to handle more complex arrangements, we’ll see increasingly sophisticated pricing that better matches the unique value propositions of AI services, including tiered pricing, regional adjustments, and more flexible contracting terms.

The future of consulting isn’t extinction but transformation, as AI struggles with proactive information gathering: While deep research tools will replace entry-level analysis tasks traditionally handled by junior consultants, the core consulting function requires capabilities AI still lacks. Specifically, AI needs to become proficient at asking good questions, managing projects, and packaging deliverables in ways familiar to the market. The most valuable consulting work involves extracting ‘frontier knowledge’ locked in people’s minds within organizations – information that isn’t available online or in databases. This requires proactive questioning and relationship management that current AI can’t replicate. Rather than eliminating consulting, AI may increase the premium on high-quality human experiences and interactions while automating routine information gathering and analysis.

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Guest Background

Max (00:01): Welcome to the next episode of the Chris Rodman show. Today we have a very special guest with us - someone with deep expertise in NLP who will also talk to us about voice AI. We’re very excited to have you, Andrej.

Andrej (00:22): Thank you so much for having me. Very excited for this episode.

Max (00:24): Awesome. Rod and I chatted about this yesterday and tried to understand what’s happening in the AI world. Your expertise will really help us understand better. For today’s episode, we’ll talk about what’s happening in voice AI, spend some time discussing AI and pricing, touch on Stripe’s recent announcement with all the AI activity within the company, and close with some cautionary tales about what deep research can and cannot do currently. Let’s start with a quick introduction about your background, Andrej.

Andrej (01:17): Sure. My background is originally in natural language processing. I did a PhD in the field pre-LLMs. Back then, we were still grappling with NLP subtasks. I like to think of it as a lot of highlighting words, connecting words, translation, etc. These were all subfields in natural language processing before LLMs really unified the field.

I worked in a couple of London-based startups, spent some time applying natural language processing at BlackRock. During the pandemic, I got into Y Combinator and started a pricing startup to help companies optimize their prices. More recently, I’m working on a voice AI product related to market research – speaking to people to gather information from them. That’s my background and story in short.

The Evolution of NLP and Generative AI

Max (02:23): That’s super interesting. Let’s start with your background. You mentioned something quite interesting to me – how generative AI has unified the NLP space. I wonder if you could elaborate on that. When you say “unify,” how does it do it, from the model itself? It’s quite interesting from a practitioner’s perspective.

Andrej (02:51): Sure. It used to be the case that natural language processing and machine learning in general was split into more manageable subtasks. Specifically in natural language processing, let me highlight three examples:

One would be recognizing named entities in a piece of text. For example, “the United Nations” is a named entity, and one of the use cases of recognizing named entities is that you can take them for further analysis – like determining the sentiment toward the United Nations in a piece of text.

That leads me to the second subtask, which might be sentiment analysis. Is a piece of text positive about a particular entity, or negative?

Maybe a third subtask would be translation – how do you go from English into German or vice versa?

If we double-click on those three examples: the first example, recognizing “the United Nations,” is almost a word highlighting problem. The second problem, sentiment analysis, is either a classification problem (positive or negative) or perhaps a regression problem if you’re trying to give a score between zero and one. And finally, with translation – that’s actually where historically a lot of generative AI and NLP draws its lineage from.

When you’re doing translation, you’re actually moving entities from one side to the other, making sure the sentiment of the text in English is the same as the sentiment in German. It’s a much more general problem. Translation was a very exciting precursor in NLP to modern generative models.

But what these modern models have done is remarkable – now you can go into ChatGPT and give it a piece of text, and it’ll recognize the United Nations for you, tell you the sentiment, or translate it into German. That’s what I mean by moving from small subtasks to a much more general model.

Max (05:34): Understood. That’s quite interesting. It’s almost like if I’m a human – I’m able to speak, eat, listen – they’re all different tasks, and then you combine it all together.

Andrej (05:46): And we’re seeing this story repeat itself. It’s a frequent, repeating historical pattern. We’ve pushed the limits of LLMs now, and we’ve moved from pre-training to test-time compute and multi-step reasoning, but maybe even that isn’t good enough. So we need to wrap it in layers of agentic processes and perhaps even layers of software. The hope is that maybe we’ll shake those shackles eventually as well.

Max (06:20): Interesting – it’s almost like you break it down to such small tasks, then build it back up and start combining with other elements to make even bigger tasks. It reminds me of the atomic level of adding different items to get to bigger things. That’s super cool. Thank you, that’s very helpful.

The Future of Voice AI

Max (06:20 continued): One of the things I wanted to check in on is around voice AI. Last year, if I remember correctly, a lot of larger venture firms in the valley had companies coming up with different types of voice AI. I wanted to get your thoughts on where you think voice AI is going based on your experience, understand how far we’ve come with voice AI, and where we are in the cycle. Perhaps we can do a little future-gazing.

Andrej (07:25): For sure. The TLDR of where we are now with voice AI is that we’re on the exponential up-ramp of voice AI getting really good. It reminds me of video calling, which only really took off with services like Skype, but you could orchestrate a video call in the 1980s – there were ways. Television networks were doing it when they had live news presenters in the field.

I remember speaking to the founder of Deepgram, which is famous for its speech-to-text models. He told this story about how in the early 2000s during his physics PhD, they created a voice AI model they could speak to in their lab. It wasn’t particularly good and was slow, and there weren’t many use cases for it back then.

But now, truly, there have been several big unlocks. First is large language models – the ability to have a general model that can handle the long tail of idiosyncrasies when people speak to voice AI. Names can be spelled in different ways, booking a table in a restaurant can be said in innumerably many ways. LLMs are very good at processing that.

The other unlock is tool calling – the ability for LLMs to call tools to action something and return it to the user in spoken form. Finally, the tech has just gotten really good in terms of speech-to-text models, text-to-speech models, and voice realism has improved massively. We now even have sound-in, sound-out models that aren’t as pipelined as previously.

But there are still bumps in the road. I expect in the next 12 months, those will be ironed out. There was interesting news just the other day about a new model from Sesame AI, which is pushing the boundary of reaching a voice AI model that you actually enjoy speaking to.

I think the crux is that speaking to an AI is actually pretty annoying. Sometimes it’s fun, but as soon as it makes its first mistake, you kind of drop it. I think we will hit a point where that isn’t the case anymore.

Max (10:18): That’s fascinating.

Voice Interfaces: Past Challenges and Adoption

Rod (10:40): Taking a step back, you mentioned that we’ve had this technology for decades. In the 80s and 90s, it was clunky and hard to use, and in recent years, it has gotten much better. However, we don’t really see that much voice as an interface. We tried a few years ago with Siri and all these assistants, and they didn’t catch on as expected. Do you have any thoughts on why that might be? Is it because we humans prefer to communicate some other way, or was the technology non-responsive, clunky, or ineffective?

Andrej (11:22): For sure. I would bifurcate this into business use cases and consumer use cases.

On the business side, the problem with voice AI historically has been achieving high enough accuracy to make it palatable for business use. Take the London-based company Poly AI, which works with large enterprises to automate customer support calls. One of their big use cases is bookings – for restaurants, pubs, gyms, etc. As I mentioned, there’s a really long tail of ways to book a place. You might call with a unique name for the local UK context, or say “I need a table for four, but there’s a child coming so I need a children’s seat.”

The way you tackled that pre-LLM was by building Byzantine pipelines to handle all possibilities. Now, the ease with which you can get extremely good accuracies has suddenly improved. On the business side, we’re seeing a large amount of adoption for low-stakes calls. What are low-stakes calls? For example, debt collection calls where you just need to continuously remind someone to pay, or slightly higher stakes like routing in freight, where a driver may call a centralized agent without having to speak to a human.

There are cases where the stakes are too high for voice AI to work well. In outbound sales, for instance, we’re seeing a spike in demand testing where people automate cold calls to gauge demand. But I don’t think we’re seeing actual applications of outbound sales calls by large enterprises trying to close sales – that’s seen as too high-stakes.

On the business side, there’s lots of application and traction already. On the consumer side, I think the biggest blocker is simply the believability of the voice. I encourage anyone listening to Google “Sesame AI” and check out probably the latest, best voice AI model available. It’s very, very good.

By believability, I mean that when we speak, the tone of our voice adapts to the current context. I’m not speaking to you in an ASMR voice now. I could convey the same information holding my microphone close and speaking in a sexy voice, but that would defeat the purpose of how podcasts are typically done – it would be a mismatch.

There’s often a mismatch between how a voice AI speaks to you – some “horrendous, agreeable, hyperactive American” voice – and your tired 11 PM self that wants to chat about whatever. On the consumer side, I think it’s much more about believability, and I think we’ll get there soon.

Rod (15:46): I just want to say that one thing I feel, especially on the consumer side, is that we’re already conditioned by decades of very limited systems. For my case, when I’m on the phone and notice I’m talking to a system, I’m always wondering what’s actually possible. I’m very cautious about what to say, wondering if the system will understand “yes” or “no.” I feel that maybe we’re used to ineffective, clunky systems, and therefore we’re very cautious about what to say.

Flow States and AI Interactions

Andrej (16:32): Yes, and there’s a parallel to be drawn with how coders used to feel or still feel. When coding, you’re typically conditioned to be very diligent – make sure the code compiles, runs correctly, has unit tests, and is properly architected. Now we’re moving into a world of “vibe coding.”

People are using tools like Cursor, Anthropic Code, or Claude Code, giving simple instructions: “do this, do that, next, next, next” until whatever compiles and runs is what they wanted. It’s more creative, more artistic.

Another good example is Adobe’s Firefly, which has incredible UX patterns that induce a flow state. One thing AI has unlocked is the ease with which we enter a flow state. We can discuss complex topics without having to walk to a bookshelf and open a calculus book to refresh our memory. It’s lowered the barrier to entry into a flow state, and voice AI is probably one of the last technologies to reach that point.

Max (18:32): That’s very interesting. When you talk about lowering barriers to flow, it’s almost like bringing the “natural” part back to natural language processing, making it easier to start tasks and get into that flow state.

Business Implementation of Voice AI

Max (18:32 continued): One thing you mentioned earlier is how businesses are outsourcing lower-stakes tasks to AI. For our listeners, especially business users, do you have a framework they can use to think about how they should implement voice AI? What would you consider “low stakes” for them?

Andrej (19:22): I think the most natural approach is to look at your existing call volumes of any sort – inbound, outbound, or internal. In those cases, you should ask yourself: is this automatable, or is a portion of it automatable?

Generally, what’s automatable is a function of risk, which is specific to your particular use case. You should ground that assessment in your context.

There are also interesting voice AI applications beyond existing call volume. With the barrier to making calls drastically lowered by AI, consider where you could leverage voice AI to improve information gathering or customer/prospect touchpoints.

For example, if you’re running an e-commerce business and an existing customer has added something to their basket but not checked out, would it make sense to give them a quick call to ask if they’re still interested? Or perhaps you want to gather feedback on their experience with a short 30-second conversation.

You’re now automating speaking to your customer in ways that were previously too expensive with human agents. So either look at existing call volume to see if you can automate a portion, or apply voice AI to entirely new use cases in your context.

Selecting Voice AI Systems for Enterprise

Rod (21:37): Thinking about that – many large organizations have call centers and customer support. I’m sure many are thinking about implementing voice systems to automate these processes. You mentioned the technology is getting quite good, with many new models in the market. How should companies decide which system to buy or which vendor to choose? Is there a checklist or best practices for making that decision?

Andrej (22:14): For sure. It depends on the size of your business. If you have existing call centers, the first question is whether it’s a BPO-style outsourced call center or something you already have in-house. Have you built some of your own infrastructure, or not?

You can divide checklists into two categories. There are the usual procurement checklists that vendors need to go through – I won’t comment on those because everyone has them. The AI-specific ones are perhaps more interesting.

One key metric to assess is the end-to-end resolution rate that a system can provide. In customer support specifically (though this can be generalized to other applications), what counts is complete resolution – not just the AI saying hello at the first touchpoint and then breaking down, requiring a handover to a human. That was one of the big problems in the 2010s with AI companies trying to automate customer support.

They would get you 60-80% of the way there, but not 100%, often introducing complexity into the flow rather than solving problems. There are other metrics that CX professionals track, but figuring out the end-to-end resolution capabilities of whatever AI system you’re adopting is crucial.

Also important is the impact distribution across use cases. It’s easy to hack end-to-end resolution scores if you only consider simple cases like “when are your opening times?” The real question is: is it actually automating the things you need most help with?

Max (24:55): The end-to-end resolution point is interesting. I can see this happening in gyms, for instance. If you attend classes once or twice, they want to upsell you to a yearly membership. Normally, staff at the front desk would call you, but you could automate that process.

AI and Pricing: Business Models and Trends

Max (24:55 continued): I want to shift gears to talking about AI and pricing. I think about this in two ways: first, how AI and pricing will change the way we price things; second, how AI companies are pricing themselves. There are stats showing companies growing from 1 to 100 million in record speed.

When it comes to revenue, it’s pricing multiplied by volume. Is this rapid growth because of huge volume or because they’re charging more for products? I’d love to get your thoughts, Andrej, given that you run AI companies. Let’s start with how AI companies are pricing themselves and reaching incredible revenue in a short time.

Andrej (26:36): I think the short timeframe, disregarding pricing for a second, is a function of three things. One is the value – undoubtedly, AI is providing value in many cases (not all, but many). The second factor is timing – we’re in an era where the tech industry’s total addressable market is higher than before.

When you see metrics on Twitter about how fast AI companies are growing, they’re often compared to SaaS companies started years ago when the market was smaller and dynamics were different. That’s hard to separate in business analysis.

The third factor, which is fascinating about the pace of adoption, is that this is one of those big secular changes. We saw something similar when the world moved to cloud computing, or when everyone became obsessed with data scientists. We’ve kind of forgotten about “big data,” but this is another wave – we all need to do AI now.

There are countless middle managers focusing on AI initiatives. I remember when I was at BlackRock, the world suddenly shifted from no one having heard of ESG to everyone doing ESG. You could tell because you’d sit in London cafes and within weeks, people in suits were talking about ESG. This is another one of those moments – for better or worse, we’re all in this together, and everyone feels they need AI. That’s what’s driving the industry growth.

Does Every Company Need AI?

Rod (29:10): On that matter, you mentioned that now we need to do AI, like before we needed to do data science. What’s the current state of play? We have those three somewhat tangential disciplines – data science, machine learning, and AI. Does every company need to be doing AI? Do they need to be doing data science or machine learning?

Andrej (29:39): No, I don’t think so. There are plenty of companies and use cases where you don’t need to be doing AI. If anything, I think the anti-cyclical play here, or what I expect to happen, is that there will be a premium on the human experience moving forward.

If AI takes over many functions and automates low-level tasks, then the human experience component becomes more important. You don’t need AI for your Pilates studio. Some of the SaaS tools you use might have AI components that are tangentially helpful, but what you’re optimizing for is the human experience of a well-led Pilates session.

Perhaps human-to-human support – white-glove style support – will become more important in a world where low-level FAQ-style questions are handled by AI.

One industry problem is that when these big shifts happen, there’s a proliferation of superficial projects led by middle managers in large enterprises who are angling for promotions. It’s important to step back and think about your company or industry as a whole, not getting too obsessed with whatever the current trend is.

Max (31:48): Absolutely. Everyone could use a little more research – as the crypto world says, “do your own research” before jumping into anything.

How AI Is Changing Pricing Models

Max (31:48 continued): Coming back to pricing, you’ve worked on pricing in AI. I’ve read online – I think it was Bain or BCG – about how AI could change pricing to make it flexible enough to respond to market conditions. I’d like to hear your thesis on how AI will affect pricing of different products and services going forward.

Andrej (32:31): I think one of the big limiting factors to pricing flexibility is actually billing, because billing is how you collect whatever you’ve priced. You cannot price something in a way your billing solution doesn’t support.

One thing I’m excited about is the unleashing of billing solution flexibility. For example, Stripe, at its core a payments company, is now also a billing company with Stripe Billing. Stripe was actually late to the billing game – companies like Zuora pioneered enterprise subscription billing, while companies like Recurly and ChargeBee handled it at smaller scales.

The flexibility of billing really limits how you price. One big talking point now is outcome-based pricing – the idea that as we move from “software as a service” to “services as software” (where AI automates what consultants do), you’ll be judged on the outcome or deliverable provided to the buyer.

What’s fascinating about outcome-based pricing is that it’s always somewhat a matter of interpretation. The outcome could be unsatisfactory, leading to disputes. If the world moves in that direction (though I don’t necessarily think it will), dispute management for delivered products or work will become really important for billing solutions.

Stepping back and bifurcating into B2B and B2C: On the B2C side, what’s fascinating is that the world has somehow decided $20 a month is the price point for AI services. I’m not sure why, but it is. Now with test-time compute and multi-step reasoning, we’re seeing higher price points because the costs of providing these services are higher.

On the consumer side, we’ll see more pricing maturity. Companies will go from flat $20/month to good-better-best models (which many already have, like $20 for basic, $200 for pro). We’ll see add-ons and regional pricing – tech pricing often starts US-biased before companies realize it makes sense to discount in certain geographies. But that’s not specific to AI; it’s a natural progression.

On the business side, what’s fascinating is you’re selling units of compute, so you naturally need a usage-based billing model. That’s exactly how AI companies have been pricing – with committed usage bands and volume discounts. Again, none of this is unique to AI, but that’s how these services are being priced.

What’s also interesting is the ratio between the actual cost of running these services, which is dramatically falling, and the price – but that takes us a bit away from the pricing conversation.

Max (37:36): I totally agree. I used to work for an old software company that sells to banks. One big reason we couldn’t change pricing was exactly as you said – our billing system couldn’t handle it. We had top software used by major banks, but we couldn’t change pricing not because clients wouldn’t pay, but because we had no idea how to bill it! It left quite an impression on me.

Andrej (38:10): Yes, and two things will happen: First, billing primitives will become more flexible. Stripe Billing and other providers will mature, making it possible to set up complex invoices with subscriptions, tiered usage, and discounts that can be edited when contracts are renegotiated.

Second, as primitives improve, there’s the question of how we’ll use this flexibility. Pricing is about matching what you’re selling to the value you provide – that’s the fairest setup in business relationships. When billing is no longer an obstacle, we may not immediately grasp how creative we can get.

Some companies have built highly flexible internal billing systems that allow their sales representatives tremendous flexibility in contracting. Without needing a deal desk to structure complex arrangements for high-value items, a seller of point-of-sale terminals to restaurants could craft custom deals – like “three months free, then discounted rates for six months, then full price by year-end, with refunds if SLAs aren’t met.”

These complex contracting situations were typically limited to high-value enterprise deals requiring deal desk involvement. We’re moving into a world where regular sales reps will be able to close deals with the same flexibility previously reserved for major contracts.

Challenges of Outcome-Based Pricing in AI

Rod (41:05): We’re moving towards outcome-based pricing. My experience with AI systems is that they often don’t get it right the first time. A classic case for outcome-based pricing might be with CRM: “I need AI expert leads based in London with specific characteristics.” Often, the results aren’t right.

However, the vendor has already incurred the computational cost of calling the model. For the vendor, the cost is already there, but the outcome was wrong. What can vendors and AI buyers do in cases where either the buyer is paying for incorrect results, or the vendor needs multiple attempts to get the right outcome before they can charge?

Andrej (42:07): Broadly two approaches: One is to say, “Look, there are pass-through costs here – the cost of using foundational models – which is the cost of providing the service. On top of that, I’m adding a margin. If this goes south and doesn’t work, at least pay for the pass-through costs.”

Obviously, customers might resist even that. In software, we lived in a world of “build once, sell a million times” where the actual cost of running SaaS was very low. Now we live in a world where margins are lower – at least for now – because the cost of goods sold is pretty high in AI.

In such cases, take a lesson from service businesses, which are incredibly good at setting expectations. Unlike software businesses where things sometimes don’t work but it’s manageable if customers cancel, service businesses need to be upfront about costs and expectations.

So either be very transparent about the pass-through costs and your role as a wrapper around existing foundational models, or get extremely good at setting expectations when contracting and anticipate situations where things don’t go well. In software, we did this with SLAs – service level agreements would define what happens if SLAs aren’t met, sometimes crediting customers to offset costs when services underperformed. We can take inspiration from that approach for AI products.

Max (44:30): I’m reminded of investment banking, where they charge based on capital raised for companies. You make more money that way, but you’re also taking on more risk because there will be days with fixed costs but little or no revenue. I like your thinking about passing on the base cost, similar to consulting where they charge for hours plus extras. Those are all service-based business models.

Stripe’s AI Strategy and Payment Innovation

Max (45:21): In our last few minutes, let’s cover a couple more topics. Stripe recently announced their latest results, with total volume reaching 1.4 trillion. Their goal has always been to grow the GDP of the internet. What caught my eye was that in their announcement, they highlighted four different AI initiatives and one thing on stablecoin. Given that we’ve discussed billing, what are your thoughts on how Stripe is tackling the AI world? They’re now one of the largest billing platforms, used by 300,000 companies with nearly 200 million active subscriptions.

Andrej (46:30): A few scattered thoughts on the Stripe annual letter: Stripe isn’t public but publishes an annual letter with high-level metrics. What’s interesting is that Stripe’s nearest competitor is Adyen, which typically services enterprises and won’t speak to you unless you’re doing significant annual revenue. Stripe’s bread and butter is getting you at the very beginning – Stripe Atlas for incorporation, Stripe Billing for YC companies.

Sometimes companies shift away from Stripe as things get more complex. Stripe’s goal has been to mature their billing systems and grow international payment coverage to keep up with larger customers, of which they have many.

One thing that struck me was the valuation multiples – Stripe is now valued much more highly than Adyen despite comparable results. That could be private market premiums, but it’s also a reflection of Stripe doubling down on the AI billing space. Most AI companies use Stripe, not other providers.

If AI takes over the world and we have AI agents shopping on our behalf, Stripe is extremely well-positioned to grow in that space. Another thing they don’t highlight much in the letter but is visible is Stripe Links – when you pay for OpenAI as a consumer, you’re setting up a small consumer Stripe account, tying your payment card for easy reuse in other Stripe situations. That’s the beginning of something potentially incredible for Stripe. Their annual letter is quite revealing in terms of strategy and direction.

Max (49:23): Absolutely. One thing I noticed is their agentic payment system, which ties to what you mentioned – if my payment details can follow me across websites, saving me from typing them repeatedly while merchants don’t store my information because it’s controlled by my agent. John Collison gave an example on a podcast about telling your agent to “go buy me some t-shirts.”

This will change how commerce works, making it more natural and lowering barriers to both flow state and low-level tasks – you just don’t need to worry about them. Combined with what they’re doing with stablecoin for cross-border payments, it’s quite exciting. From your perspective, building in the AI and market research space, how do you think about this? They’ll eventually be able to see where the world economy is heading as everything moves online.

Andrej (50:42): Yes, I caught either Patrick or John on a podcast talking about how they developed an inflation prediction model, which is pretty cool. There are many interesting possibilities assuming the world fully moves toward programmatic payments – easily programmable payments.

Just the other day, I was reflecting on something from my background – I’m half Slovenian, half Serbian, from ex-Yugoslavia. Companies there would submit their debts to a government agency regularly, and that agency would cancel out circular debts so that financial actors wouldn’t take commissions on those transactions. In a world of easily programmable payments, many interesting things could happen.

Deep Research: Capabilities and Limitations

Max (52:04): Great. In our final minute – Ben Evans, the former A16Z investor and prolific researcher, has been giving us reality checks about what AI can and cannot do. He recently wrote an article about deep research capabilities and limitations.

He noted that “AI models are very good at things computers can’t do well, but very bad at what computers do very well.” Computers excel at deterministic tasks, while AI is better at probabilistic tasks like pulling information from multiple sources. But when you need something very specific, AI becomes more unreliable, though it will improve. I’d love your thoughts on this, given your journey from relating different entities to seeing those subtasks combined.

Andrej (53:23): It’s true that AI excels at probabilistic tasks and struggles with more deterministic computer-like operations. However, I’d argue that you can always adopt intermediate representations to handle deterministic tasks. AI is incredible at coding, which translates into deterministic actions across data.

For example, if you’re doing data analysis with SQL queries, you can either give entire spreadsheets to an AI for immediate answers, or it can generate code that runs across your table to produce the answer. The trajectory of improvement in this area has been amazing – in the 2010s, people tried to generate SQL code with little success, but now it works well.

Regarding Benedict Evans’ point, one thing that’s often conflated with deep research – not by Ben himself, but by many on Twitter – is the idea that “this is the end of consulting.” I take major issue with this because it misunderstands what consulting is about.

It’s not just about middle managers needing justification for projects and bringing in McKinsey – that’s a cynical view. What consulting often involves is cases where an AI would need to be proactive in pulling out information.

I think about knowledge in tiers: there’s online public knowledge on the internet that’s scrapable; there’s near-field public knowledge available through subscriptions like Gartner reports; and there’s frontier knowledge locked in people’s heads and organizations.

Consultants often speak to internal stakeholders about how they’re tackling problems – essentially pulling frontier knowledge from their minds. To replicate that, you need an AI that’s very good at asking questions, and we don’t have that yet.

For anyone thinking about automating consulting, you need an AI that’s good at asking questions, managing projects (timelines, expectations), and packaging deliverables in ways familiar to the market. Consultancies have knowledge management teams with pre-existing decks, solutions, and guides for their consultants because they’re often repeating similar work across different clients. Taking inspiration from that approach makes more sense than just searching across Google.

Rod (58:02): One thing I find is that with deep research products like Perplexity or Gemini, they seem to replace what would often be given to the intern or entry-level analyst. While consulting might not disappear as an industry because of brand power and prestige, is it possible that many consulting-related roles will completely disappear?

Andrej (58:53): Absolutely. We’re moving into a world where low-level legal services can be answered by ChatGPT, medical scans can be analyzed by AI, and secondary research can be done using deep research tools.

But going back to Jevons paradox (which everyone started talking about a few weeks ago), what often happens is proliferation rather than replacement. More people engage with research because it’s more accessible. Someone on Twitter mentioned we need a flow state-inducing AI that your “lizard brain” uses before bed instead of scrolling through TikTok or Instagram. Someone else replied that they’re already doing this – hunched over their pillow chatting with ChatGPT about research while their partner watches TV.

So AI will definitely increase the aggregate amount of research, which is awesome.

Max (01:00:35): Yes, Jevons paradox – Satya Nadella mentions it too. Increasing the efficiency of resources can lead to increased consumption of that resource. When we improve research capabilities, we’ll likely just consume more of it.

One positive outcome might be that investment bankers and consultants, especially junior analysts, will have better work-life balance. In services businesses, the higher you move, life doesn’t actually get easier. It gets harder because you have to spot-check all the work your teams are doing, while still managing client relationships and asking the right questions.

I think there’s an opportunity with voice AI – you could easily call someone using voice AI, ask a few questions, and capture enough information to form an initial opinion before going deeper. I started my career asking people questions like “would you use this software?” which I think will change. I’m quite excited about it.

Andrej (01:01:58): Just to add one thing: Another fascinating topic to think about is what kind of information we’ll be extracting from experts in the future. If folks in an enterprise or industry have used services like Tegus, these are like a window into the future – they have open transcript libraries you can subscribe to. A lot of that data is very interesting and will have enormous potential when combined with AI.

Max (01:02:46): Yeah, and the question then becomes who owns that data and so on. We’ll get to that when the time comes. But I totally agree – there’s still a lot to unlock, especially with nuanced information in large organizations. I can only imagine what we’d find if you combined all the transcripts and internal filings of IBM.

Andrej (01:03:10): Mmm-hmm.

Max (01:03:13): Just imagine what you would see there. You could say the same for Microsoft, Google, and others. First off, Andrej, thank you so much for giving us extra time. We really appreciate it. This was a very good conversation. To wrap up – thank you again for sharing your expertise with us. Today we covered four different topics: voice AI, AI pricing, Stripe’s recent announcement, and what deep research can and cannot do, as well as the future of consulting.

Thank you everyone for tuning in and listening. If you like the episode, please share and subscribe. And Andrej, if people want to find out more about you, where can they go?

Andrej (01:04:03): Gosh, I guess two venues. One is, add me on LinkedIn. The other is come to London, Bermondsey Street, Watch House. I’m there all the time and you can spot me in the wild.

Max (01:04:15): Lovely. There you go, listeners. You know where to find Andrej. Thank you so much again for listening, and we’ll be back next week. Thank you.

Your Hosts

Christine Wang: https://www.linkedin.com/in/christinewang0/ 

Rod Rivera: https://www.linkedin.com/in/rodriveracom/ 

Maxson J.Y. Tee: https://www.linkedin.com/in/maxsontjy/ 

Social: X: 

https://x.com/chrisrodmax 

Instagram: https://instagram.com/chrisrodmax 

LinkedIn: https://linkedin.com/company/chrisrodmax 

Subscribe: https://chrisrodmax.com 

Subscribe now

Rod Rivera speaks with Gus Neate, CEO and co-founder of Wilson AI, about the transformative impact of AI on the legal industry.

They discuss Gus’s unique background in engineering and law, the capabilities of Wilson AI in automating legal processes, and how it enhances efficiency in contract review and legal inquiries.

The conversation also addresses concerns about job displacement in the legal field, the importance of security and confidentiality, and the potential for Wilson AI to support small organizations without in-house legal teams.

The episode concludes with insights into the future of legal work and the integration of AI into existing workflows. In this conversation, Gus Neate discusses the integration of Wilson, a legal tech tool, with existing tech stacks, emphasizing the importance of security and customer needs.

He addresses concerns about AI hallucinations in legal contexts and explores Wilson’s capabilities in legal research and contract drafting. The conversation also covers Wilson’s unique selling proposition, trends in legal operations, and advice for early career lawyers in adapting to an AI-driven legal landscape.

Neate highlights the evolution of AI in legal tools and provides insights on how to get started with Wilson.

Chapters

• 00:00 Introduction to Wilson AI and Its Founder

• 06:01 Streamlining Contract Review Processes

• 09:07 Customizing Wilson AI for Corporate Needs

• 12:09 Addressing Concerns of Legal Professionals

• 15:07 The Role of Wilson AI in Small Organizations

• 18:05 Security and Confidentiality in Legal Tech

• 20:51 The Future of Legal Work with AI

• 23:54 Integrating Wilson AI into Existing Workflows

• 25:45 Understanding Customer Needs and Security Concerns

• 27:37 Addressing AI Hallucinations in Legal Contexts

• 29:50 Exploring Legal Research Capabilities

• 30:39 Current Product Roadmap and Future Focus

• 31:58 Enhancing Contract Drafting with Wilson

• 33:03 Wilson’s Unique Selling Proposition

• 34:52 The Role of Templates in Legal Processes

• 35:35 Trends in Legal Operations and AI

• 36:45 Advice for Early Career Lawyers

• 44:04 Future-Proofing Legal Careers in an AI World

• 45:19 The Evolution of AI in Legal Tools

• 47:11 Getting Started with Wilson

Takeaways

Wilson AI transforms legal workflows by providing instant answers and automated contract reviews: Wilson AI functions as an AI paralegal that integrates with tools like Slack to answer legal questions in seconds rather than the hours or days it might take busy legal teams. The platform automates contract reviews using customizable playbooks, flagging concerns with a red/amber/green system that helps non-legal teams quickly understand risk levels. By training on organization-specific data, Wilson provides contextually relevant answers based on a company’s existing policies and documents. This approach dramatically speeds up routine legal processes, allowing companies to move faster while maintaining appropriate risk management.

AI augments rather than replaces legal professionals, freeing them for higher-value work: A key insight from the conversation is that AI tools like Wilson aren’t designed to replace lawyers but to enhance their capabilities. By handling repetitive, routine tasks that often distract legal teams from more important work, Wilson enables lawyers to focus on strategic initiatives and complex matters that require human judgment. This augmentation allows legal teams to handle more work and support more clients at a higher level. Additionally, Wilson provides valuable data insights that help legal teams better understand workloads, identify training needs, and demonstrate their value to the C-suite. Far from threatening legal careers, AI tools are allowing lawyers to work more efficiently and deliver greater business impact.

Future legal professionals should embrace AI tools as part of their skill set: For aspiring lawyers or early-career legal professionals concerned about AI’s impact on their career prospects, Gus offers clear advice: law remains a fantastic and promising career path, but professionals should stay close to AI tools and practice using them. Understanding how these technologies can enhance legal work is becoming increasingly valuable in both law firms and in-house teams. Rather than fearing automation, today’s legal professionals should embrace these tools and potentially position themselves as drivers of AI adoption within their organizations. Those who can bridge the technical and legal worlds will be particularly well-positioned for success, similar to Gus’s own path from engineering to law and eventually to legal tech entrepreneurship.

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Welcome

Rod: Welcome to another episode of the Chris Rod Max show, where every week we discuss the latest news, developments, and what's happening in enterprise AI. Before we start with our show of the day with our guest, I want to remind you to subscribe to our newsletter at chrisrodmax.com. And of course, like, subscribe, and leave your comments under this video. Today I'm joined by Gus Neate. Gus is the CEO and co-founder of Wilson AI. Hi, Gus. Great to have you here today.

Gus Neate: Hi, Rod. Thanks so much for having me on. Really looking forward to speaking with you.

Gus's Background and Journey

Rod: Before we get started about what Wilson AI is or the space it operates in, can you tell us a little bit more about yourself? How did you come to be building AI today, and what led you to the problem you're solving?

Gus Neate: Of course. To give you a bit of context on myself, my background is that I originally studied engineering at Oxford. I was technical by background. Most of my studies focused around a combination of information engineering systems and also entrepreneurship, which I was really interested in.

Towards the end of university, I entered an essay competition run by Clifford Chance, one of the largest law firms in Europe. Off the back of that essay competition, they offered me a place to join the firm. So I went through law school and actually became a lawyer, which is a very unusual route for an engineer.

Having become a lawyer, I saw how many areas within legal practice still have potential for automation. It's actually remarkable, and I'm happy to dive into more of those. Particularly since the developments of large language models and the popularization of those—especially the release of GPT 3.5 about two years ago—the amount that can be automated has really changed. That's what inspired the shift.

In the long term, I think there's potential here for massively improved services within businesses and massive increases in the quality and speed of responses you can get when you have legal challenges.

Rod: Curiously, I know some profiles similar to yours—people with technical backgrounds, engineers and so on, who decide to specialize in law, especially in the area of patents. That field requires very in-depth technical knowledge, and as a result, it's not just possible to be a "pure-blooded" lawyer—one needs to have this technical background.

Gus Neate: Yes, I think you're right. One thing I would say is that software engineering and law are both super technical, very detail-oriented, and actually very precise. A small change in language in both a contract or code can make a massive difference. So the skills that you use in either engineering or law are often surprisingly similar—very technical people looking for risks as well, often. So yes, I agree.

What is Wilson AI?

Rod: If I go to the Wilson AI website, I see "Legal has never been so fast." Why is that? What is happening?

Gus Neate: Absolutely. To give a bit of context, what happens at the moment in many teams is that they'll reach out to legal, and legal teams—especially in corporate settings—are often really overwhelmed. There's a huge amount of work across a really large variety that comes through those teams.

What's interesting is that by deploying Wilson, which trains on your data, people get extremely fast responses based on their existing data that can then be triaged into the legal team if it's more complicated.

To be more specific, we have an integration, for example, into Slack, where people can ask questions that they might normally direct to the legal team. Wilson can often immediately answer those questions. That takes five to six seconds usually to produce the answer. Whereas naturally, if your legal team is even up to 20 people in a large company, they'll often be in meetings or busy reviewing documents, and they can't break their flow state, which means your responses will be much slower through the human process.

You see the same thing across other workflows, like reviewing contracts. Currently, for many teams—and this is how I used to review contracts—I would usually be running through with a playbook, which would have all the key points that I needed to assess. Then I would manually go through, identify where they were, and then think very carefully about whether language needs to be changed or if there are risks.

But our tool can automate that process. Wilson will identify the risks, suggest the changes, and then lawyers jump in and it's as if someone has already been through and done the first part of the document. They just need to go and check through the outputs.

Practical Use Case: Contract Review

Rod: That example of contract review is something I can identify with. When I think about situations in my day-to-day office life where I need to request advice from a legal team, it's often when we want to start a partnership with someone or buy a tool. Then we have an agreement and the first question is: "Can we sign it? Does it have everything we require? Is the contract correct?" I'll then try to find someone who can help me. So if I'm in that situation and I go to Wilson, what happens? How will it help me?

Gus Neate: Exactly. You're completely right. For both vendor contracts or partnership agreements, those can be put into Wilson. You can load the document directly into Wilson, and you can either use out-of-the-box review playbooks—which would be using the positions that I've seen from my in-house background, and we've researched and tested at Wilson AI—or customize them.

When you load the document into Wilson, it will run through and analyze every point of the document to see whether it's off market or doesn't align with the points that you need to review. We do it with red, amber, and green flags. If there are any red flag risks, that would be a case where you would think, "Okay, this one I'm going to need to loop in legal on." But when you see a ton of green flags, you can sense, "Wow, this is close to ready to sign, no major issues." Usually then you might route that to legal for their final approval to move towards signature.

That process is done much more quickly when the first pass of reviewing the document can be automated using AI.

Customization for Enterprise

Rod: I imagine that especially at large organizations, there are so many requirements that are specific to these companies. One cannot take a vanilla approach of saying, "These are just common sense rules we should be following," but rather should be very tailored to the company specifically. How does this happen? How is the tool functioning such that if I'm a very large corporate, I can really make it fine-tuned and tailored for my own processes?

Gus Neate: You're absolutely right. There are two really big ways that we do that.

The first is when companies already have existing playbooks, they can load those into Wilson, which will then overlay those on top of the standard playbooks to adjust them to your specific positions. Those are then still also editable by your legal team. So it's super fast to set up—you load in your existing positions and then Wilson is ready to go. That's one piece around contract review playbooks.

The other piece that's pretty exciting is we've just added a number of integrations with existing data systems where people are storing documents—places like Google Drive, OneDrive, Confluence, or Notion. Those can be instantly integrated with one click, and those pieces of information pulled into what we call Wilson's Superbrain. The Superbrain effectively is what Wilson as the AI tool knows about your organization. When we do those one-click integrations, we pull information from all around your organization straight into the knowledge base of Wilson so that powers its outputs.

The last bit to mention is that Wilson can help with highlighting where things are inconsistent or allowing you to amend policies. Wilson might give snippets and citations for where it found answers. Sometimes we see teams using Wilson and finding, "Okay, it's given two citations there. I see we should be using our more up-to-date policy." They would then tweak that in their Google Drive folder, and then that automatically powers Wilson the next time for its reviews.

Rod: And for those who do not know, what is a playbook?

Gus Neate: Yes, absolutely. Sorry, it's a great point—I shouldn't get too technical. A playbook is usually a list of points in a contract, such as indemnities or termination provisions, and then it's a list of positions that you would review for or risks, and then pick points you would look at.

For example, in a vendor contract, termination and auto-renewal is a classic point that legal teams will be looking at. They're often training sales teams about this. That will be flagged, the position identified, and Wilson will assess whether it aligns with your business risk level and commercial position. That's what people would flag as a lawyer—to make sure you're getting positions that are favorable to your business. Sometimes these positions, when it comes to an exit for a business down the line, are extremely important for how their revenue is valued.

Impact on Legal Professionals

Rod: I can see how those who require legal services at the organization benefit from Wilson. But on the other side, if I'm thinking from the perspective of a lawyer, I might worry about being replaced. If many of the tasks I'm already doing, such as checking clauses and making sure everything is correct, are already automated, will there be less work for me or no work at all left?

Gus Neate: It's an interesting point and a valid concern. The way we see this is that what we've often observed is that the lawyers we're working with are able to handle a lot more work and support more clients in the business at a higher level.

The other piece I often see in legal teams is that they're really swamped. So there are two big focuses. Often they want to be able to do more of the higher-value, more strategic work—whether that's identifying risks and looking around corners in the business, or joining and setting strategic direction alongside the business team. By taking a lot of that lower-hanging fruit off their plate using AI, they're freed up to do more of that.

We're also seeing that with the in-house legal team having more capacity, sometimes that allows them to handle contracts that might have gone externally because of capacity constraints. So it might be the case that you receive a large high-value commercial contract that needs to be turned around within one day, but your internal legal team doesn't have capacity. By using Wilson, you're freed up to do more of that work internally.

Those are the two big ways it's changing things. One thing that we are very conscious of is that instead of having jobs replaced, we also provide training. We show people how they can do more work, and often that actually unlocks more things for them to do. I know there are lots of concerns around this with AI, but there were similar concerns around many technological innovations. And usually I think the way this will pan out is much more efficient and higher-quality services, and actually more work being done.

Common Use Cases

Rod: We discussed the use case of contract review. What are the most common workflows or problems that your customers are solving with Wilson?

Gus Neate: I think if I put it into three big buckets:

The first bucket is questions that come through their legal or operations team that are repetitive and generally often stored inside an internal policy. These are a distraction from higher-value work. Those are automated with Wilson.

The second bucket is around reviewing contracts, flagging key risks, and then actually marking up the documents—making changes to the language to ensure it aligns with what you need. That's often more of a legal user role, whereas the questions are more for commercial users from sales, marketing, or product teams coming to the legal team.

The final bucket is something quite interesting that many in-house teams are focused on: Wilson provides an oversight pane of everything going through the tool—all of the AI calls, but also all of the work being allocated within your legal team and metrics. We've seen a lot of teams struggling to get good metrics on what's going through the team. They want to understand how many questions they get around data privacy, how many around HR, how many around commercial contracts. Should they be training people on particular topics as they get tons of questions there? Or do they need more resources and to hire a new data privacy lawyer because there are many queries or contracts coming through that are data processing agreements?

We see both lawyers wanting more insight into this and often C-suites asking their lawyers to report on it. Wilson can plug in and provide those metrics and the oversight that teams have been looking for. That makes them a lot more risk-compliant as well because they have better oversight into everything coming through the team.

Target Industries

Rod: Are there industries or specific types of companies where you feel Wilson is the perfect tool?

Gus Neate: Absolutely. Most of our customers at the moment tend to be technology companies. We've seen that the pain points we're addressing are particularly acute in fast-growing companies. But what's interesting is these problems tend to be prevalent across organizations of different sizes and across different industries. But at the moment, we tend to focus on working with technology companies.

Support for Companies Without In-House Legal

Rod: For technology companies, one challenge I often see is not just asking if a contract is valid, but drafting one. When a company is growing and wants to close a partnership with a large organization or sell their software, especially if we're talking about small organizations, they might not even have a lawyer in place. Can Wilson support companies that do not have in-house lawyers, or do I need to have someone in the team responsible for legal?

Gus Neate: This is a great question. We are seeing some teams without in-house counsel. One thing Wilson can do, and we have some partners we work with here, is serve as your first point of call to review a contract or answer questions. But Wilson does need to be supervised by a qualified lawyer in the relevant jurisdictions for the advice.

The way we're solving this is by partnering with organizations—sometimes there are fractional general counsels or fractional general counsel networks. What that means is you can use Wilson, and if you received a commercial contract but don't have an in-house lawyer and it's a larger scale contract than you're used to, you could run it through Wilson and get the initial advice. If you want to turn that into legal advice, there are partner lawyers who can jump in and say, "I'll review that contract for you. I'll pick it up from Wilson, double-check the output, and make sure it's aligned with your positions."

So that's how we see that working. But yes, we do see teams that haven't yet got legal counsel interested in using Wilson.

Data Security and Privacy

Rod: Much confidential information is included in contracts, including attachments. There might be company secrets there. Some might have concerns about uploading to a tool and not knowing what happens with that data. How do you address those concerns for people who are fearful of revealing their company secrets to the outside world?

Gus Neate: Absolutely. Right from the beginning, we've had enterprise application in mind. All customer data is kept separate and tethered to their individual organization. We're secured with an organization called Clerk, an amazing auth provider. We've tied everything back to a specific organization, so nothing leaves that organization itself. No sensitive data is ever held on the client side—only on secure servers. And that's by design.

On the model layer, one important thing is that we use the OpenAI Enterprise API, which means no data is trained or saved by OpenAI. Those are the main things we've done to really allay those security concerns. You're absolutely right—we hear this from legal teams all the time.

Before and After Wilson

Rod: Can you describe a common setup for a company that hasn't yet adopted Wilson? What's their current status quo and day-to-day, and then what happens after they adopt the tool?

Gus Neate: Absolutely. Generally, what we would see in teams before Wilson is that they're often struggling to keep track and feeling distracted, spending more time on items like repetitive questions or looking through tons of vendor contracts or NDAs that take up much of their time. They're unclear whether this is just their team's issue, how many contracts they're really doing, how much of their time is getting distracted when people keep pinging them with questions, and how much time they could be freeing up.

Once they deploy Wilson, it handles a large proportion of the lower-value work, but they're also getting insights into exactly how much of that work exists. So the way they plan and manage their team can be quite different.

In the future, I see legal teams being a combination of AI agents and people. What Wilson really is, is an oversight layer through which you can orchestrate a combination of AI tools and the people in your organization.

Inspiration for Wilson AI

Rod: Going back to your time as a lawyer, what gave you the inspiration to start Wilson? You were working as an engineer, so you already had technical proficiency, but why leave a likely very comfortable job with a specialized niche that is very well regarded? Lawyers have always had a prestigious position, and then suddenly you decide to do a startup in the legal space.

Gus Neate: It's a great question. I think there are two really big things. One of which is that I was using AI tools and testing them on a range of flows that I was doing at work, and I was super interested to see the possibilities. I started using GPT early with 3.5—Clifford Chance also hosted a private instance, so I was able to use that. You could see the quality of the outputs and the pace at which they were improving was extremely rapid.

That gave me high conviction that in the long term, AI has enormous potential to assist in the legal space. The other piece was seeing tools like Cursor cropping up, which within software engineering has become ubiquitous. That's not so much the case for lawyers yet—there isn't really an equivalent. And I think there's an enormous potential to build systems that massively improve a legacy industry and improve the quality and pace of work.

You can see that's happening on the AI side, but there's a big responsibility in helping people adopt these tools. I think in industries like legal that are more traditional, it's super helpful to have a background where you understand the workflows and can talk people through how to adopt these tools safely and where they're best placed to use them. I was very lucky in that my background as a combination of engineering and having become a lawyer means that I'm fairly well-placed to help people understand how to adopt these tools.

The Concept of a Legal Copilot

Rod: You mentioned Cursor, which for those who don't know, is a tool for software development where traditional software developers have a copilot that speeds up the programming process. So the idea is that Wilson becomes this copilot not only for users of legal services in-house but also for the lawyers themselves—speeding up their work and helping with their daily tasks.

Gus Neate: Absolutely right. At the moment, we focus on in-house corporate legal teams. Similar to Cursor, it's really changed day-to-day work. If you're interacting with software engineering, people would say it's like night and day—they're producing 10 times as much code. In legal, it's not yet the case that people think they're doing 10 times as much work with AI tools. But I believe Wilson will enable that, and that's what we're building.

AI Adoption in Legal

Rod: You mentioned GPT 3.5 too, which makes you sound like a trailblazer in this space. What's the current situation for lawyers? How are they currently using AI in general? Are they already adopting it, or have they not even heard of GPT? What are you seeing in this space?

Gus Neate: What we're seeing, which is really interesting, is that there have been several waves of legal tech innovation, often hailed as revolutionary. And I think that's happened several times. So within the legal profession, and quite fairly, there can be a bit of fatigue around "I've been told before that natural language processing was going to revolutionize legal work."

What's interesting is we're starting to see teams using more and more tools. This time, I think they're finding that it is different—it's changing their workflows and really saving them time. So a lot of the teams I'm talking to are looking at tools more and more and finding that the level of time savings has increased very rapidly in the last year or two versus what was previously available.

There's a ton of interest. What's interesting is you sometimes see people who have tried a tool, and that's inspired them to understand what might be possible elsewhere in their organization. Similarly, if they see their software engineers deploying tools, or their infosec teams using tools, they often think, "Actually, we should be checking out these tools because they might be really valuable for our business."

Technology Adoption Among Lawyers

Rod: I'm curious about that part because I know a few lawyers, and my impression is that this is more like an offline, social profession. For example, if they have a document, they will rather print it and highlight it with markers. They'll be working with Microsoft Word, sending emails—not exactly very technologically sophisticated. And then you come here with a very advanced AI tool. Is it difficult to explain? Is it difficult for the lawyers to understand what Wilson is about and embrace it?

Gus Neate: I think it's a really interesting point. That is how the traditional side of the industry works. The way that we focus on this and think about it is it's really about UX—making the tool extremely intuitive, building it very specifically for legal use cases, and making it very easy to adopt.

One way we do that is by going directly to the tools they're already using. For example, whether that's inside Slack or inside email, if that's where you're used to seeing your outputs, Wilson will output into those tools. A lot of people are used to having someone in their team assist with this work, so Wilson can often give responses as if it's a junior member of the team. That really can help people who haven't adopted and used these tools before.

A combination of fantastic UX and the fact that we're seeing people really interested and excited about trying these tools has massively increased adoption.

Platform Integrations

Rod: Right now you have this integration with Slack, so the primary touchpoint and user interface with Wilson, as I understand it, is through Slack. However, in large organizations, they might use Microsoft Teams or they might not really have a messenger—maybe there is something in-house. What happens in those cases? Can you still support them?

Gus Neate: Yes, we can support them through the web app, so people can interface with Wilson both through our website and web app, and also through Slack—and they both sync together. When you integrate them, anything asked in Slack also populates into the web app, into the oversight pane that I was talking about.

In terms of more traditional tech stacks, like Teams and email, we're actively working on putting out those integrations too, so that Wilson can be deployed across an organization with any tech stack.

Sales Insights and Customer Discoveries

Rod: Thinking about how you are approaching enterprises and your sales process, what realizations have you had along the way? Is there something you noticed—maybe a request people had that you really didn't think about, or something you offered that people said, "Wow, we didn't know we needed this, now we need to start using it"?

Gus Neate: A couple of things come to mind. Security is, of course, one of the big ones—using enterprise security and building that right at the core has always been key. That's definitely been reinforced by conversations with our customers. It's really important to make sure their data is segregated and secure, and to ensure the foundational models aren't trained on that data.

Another more fun discovery is about usage patterns. People can ask questions to Wilson either in Slack in a public channel, or more privately through direct messaging or through the web app. One interesting discovery is that sometimes people will ask questions that they might have thought were silly or that they were reluctant to ask publicly—they like to be able to ask privately. We hadn't anticipated that would be the case. But people do like to be able to ask something that's not always in the public channel, while the lawyers still have oversight and can jump in. This opens up opportunities for people who are unsure where to get their legal advice to ask in a more secluded setting.

Addressing AI Hallucinations in Legal Context

Rod: Mentioning the wider range of use cases and the different areas that legal touches, one challenge with generative AI is always the risk of hallucinations—of AI making up answers. With legal work, one needs to stand on secure ground with certainty that whatever the system is saying is correct, so you can make decisions based on that. Are there mechanisms you have in place to prevent or counter the risk of the AI making things up?

Gus Neate: Absolutely. We're super focused on this and do a lot of testing around hallucinations. The two big pieces we do here are:

First, every time Wilson gives an answer, it sources where it found that answer from, whether that's from our internal knowledge base or yours. You would see that it found the answer in a specific policy, and it provides the snippet alongside its answer—you can click straight through. It's similar to how you may have seen tools like Perplexity doing this. Generally, we've found that trust in AI is heavily increased by making sure you can see exact sources. That also massively reduces the risk of hallucinations.

Second, we make the UX very transparent. We think it's important that Wilson is not just correct in its answer, but what we would call "very easily verifiably correct." You can both see that the answer is correct, and then you can quickly hover over and see where it got the information from and verify it's a trusted source. Or you can easily escalate to a lawyer, have them looped in to confirm they're okay with that position, and then get back to you. So a lot of what we're doing is building UX that allows AI outputs to be easily verified as accurate.

Legal Research Capabilities

Rod: You mentioned Perplexity, which for those who don't know, is a browser search engine with AI that lets you ask questions, and it crawls the web to provide a summary of results. When I think about legal processes, it's not only about existing contracts I want to review, but very often people have questions about their own personal legal situation. There's a lot of exploration where, for example, I might ask about current best practices for a vendor contract or what clauses I want to include. With Wilson, is it also possible to not only discuss existing processes and contracts, but also explore things I want to implement in the legal space?

Gus Neate: The bucket I would put what you're describing in is what we think of as legal research. What we're doing at the moment is building out increased trusted data sets that people can use so you can do your legal research through Wilson. We see teams wanting to use AI for this.

We also actually do a lot of this currently with customers. We speak with them and do research, and using my legal background, I'm able to give advice about what we're seeing in the market. But I think there's a lot of potential down the line for legal research. It's on our product roadmap, but it's not the core thing we're focused on right now.

Product Roadmap and Focus

Rod: If we now talk about the topic of product roadmap and your focus, what would you say is the current roadmap and your focus?

Gus Neate: Absolutely. As I mentioned earlier, we're increasing the number of integrations. For teams using Microsoft Teams or Word, we're building out those integrations.

Wilson already completely answers questions or reviews contracts. We're now going deeper and deeper into workflows so that Wilson can go end-to-end. Those are the two big things—we're building more integrations, and also making sure Wilson will go all the way through a workflow, whether that's integrating into sending documents out in DocuSign, or drafting approval emails, looping in additional stakeholders, those sorts of tasks.

In the longer-term product roadmap, there are quite a few pieces around legal research which are in the pipeline as well.

Assistance with Legal Language

Rod: You're also mentioning drafts. Commonly, how contracts and legal documents are written is not necessarily the language we use day-to-day. Even for a native speaker, this type of vocabulary is uncommon. Does Wilson help me if, for example, I'm drafting a contract and using common words—will it tell me, "Actually, this is not the correct term, replace it with this term that is the legal term"?

Gus Neate: Yes, exactly right. Wilson sits alongside you within, say, Google Docs if you're drafting a contract, and you can speak with Wilson. It's similar to other tools like Cursor where you can be interrogating Wilson saying, "In this clause I've written it this way, what do you think would be the language that an expert lawyer would use?" Wilson will then come back and clarify language and use the appropriate terms.

It's worth mentioning that in my view, the best drafting often is extremely simple, readable, and legible for people. There's a big focus on making sure things are clear and easily understood, rather than using traditional legal language and legalese.

Competitive Differentiation

Rod: Given the wide array of use cases and situations where Wilson can be implemented, I'm sure there are other tools on the market that either are similar or cover some areas of what Wilson is offering. What is the main sales argument or differentiator, the unique selling proposition that you have with your customers?

Gus Neate: Absolutely. One of the top things we're seeing with Wilson is that it sits as a first line of defense and a first port of call for commercial teams or business teams. What's really interesting is Wilson is often positioned between legal and the business. There's no barrier there, but what it's doing is picking up tasks and doing them extremely quickly, directly for commercial users. That's one thing we've focused on that's a little different from other companies.

Many of our users would be from a sales team that wants really fast responses to go back to a customer either on a contract or a question, or someone who has a question about their employment contract and wants quick responses to understand their entitlements. Wilson is able to sit in that gap—it's not just used by lawyers. I think that's very different.

The other thing that's quite different from what others are offering is what we sometimes call a "legal pane of glass." As a legal team, you're looking to get more oversight across your business into what's occurring, what the risks are, and to be more data-driven. What we've built in terms of the dashboard, oversight ability, and automatic tracking of everything that's happening—we haven't seen many tools on the market that are as good at keeping track across your team as what we have.

Contract Templates

Rod: You mentioned that your core focus is technology companies that are growing fast. In the legal space, there are so many routine processes for technology companies, such as hiring an employee, creating job contracts, handling terminations, and so on. For these, I imagine that AI might not even be required—it's more about having the right templates that are reviewed by good lawyers and approved. Do you have a database of contracts that companies can browse through and say, "Now I need an employment contract" and get a draft for this?

Gus Neate: Interesting. That's a great idea. We don't have a current database that people can use in that way. The way we usually see this is people tend to take their template and then ask Wilson to populate it, for example, for employment contracts, or make it specific to their needs. The drafting tool where Wilson will review and assist with amending contracts can be used to tailor your existing templates. We don't currently provide templates of documents to people, but it's a really interesting idea.

Legal Tech Trends

Rod: Thinking about the technology space and just the idea of legal ops and AI, what trends are you seeing in the space?

Gus Neate: I think one thing we're definitely seeing is that legal ops is becoming more and more core for a lot of teams. Over the last four or five years, it's interesting to see how early legal ops is becoming a key hire for in-house legal teams. You might see even the second or third hire—people would often hire a general counsel as their first hire, and then after that, the second or third hire will often be a legal ops manager.

Setting up these systems and tools is a super useful way to get really high amounts of leverage with your legal team. We often see very high ratios and increasing ratios because people are able to support more and more people around the business by using these tools. So there's big growth in legal operations as a sector and in the importance of hiring that role early and having those people in place.

Advice for Legal Hiring

Rod: You're in contact with many companies, and these companies sometimes are not in a position to have their first legal hire or are about to make one. Do you have any advice for them? What should they be looking for today in a good legal hire?

Gus Neate: Absolutely. A couple of the characteristics of a good legal hire, and this is often what you see for people moving in-house, is they really want to be close to the business and understand the business aims. I think that's extremely important.

When you're dealing with the sales team, it's all about thinking about what's the focus for the sales team. What are they looking to try and get done? The same thing with the marketing team. The best lawyers or legal operations and legal professionals are often really deeply ingrained into the business and trying to drive your core targets, whether that's a revenue-based target or a sales-based target. They're looking to effectively manage your risk, but really remembering that the key goals of the business tend to be around things like sales and driving more revenue. That's something to look out for in hires.

The other key piece is really clear communication skills. Lawyers interface with the full range of the business—generally, there's almost no area of a business that isn't touched by legal. So being able to have a high EQ and great communication skills to deal with a big range of stakeholders is super important.

Improving Communication Between Legal and Business Teams

Rod: The communication skills point is very interesting. From experience, often when talking to lawyers, they might use language that is harder to decode—they struggle formulating their thoughts outside of legal vocabulary and frameworks. I often see the lawyer might be talking in a way that for the layman is very difficult to understand. I'm wondering, could we imagine that now that we have more conversational AI, Wilson could also help with voice and discussions, and translate back and forth between very technical, legalistic discussion and common language that's easier to understand?

Gus Neate: I think that's really interesting and exciting. We do keep close tabs on multimodal AI, and we often do things like hackathons where we try out these tools. I think it's really interesting to deploy Wilson via a voice agent to be able to talk to Wilson and have it give really clear and easily understandable legal outputs. That's something we are looking into and experimenting with.

One interesting thing we've seen with teams is that the vast majority of communication with lawyers is usually written communication. Most people contact their lawyers via Slack or email or by messaging them in text, and most of the outputs lawyers provide are written down. So while it's exciting, it's not our top focus right now, but I think it's a really interesting future direction.

Geographical Reach and Jurisdiction

Rod: Another thing is that law is very country-specific. I imagine you're having customers from all around the world. Can any company use Wilson, or are you only specialized in specific geographies?

Gus Neate: Our two key markets that we're focused on at the moment are the US and the UK market. But it's worth mentioning that Wilson works across multiple languages and across multiple jurisdictions.

The thing we do recommend is that Wilson acts like an AI paralegal. That means it's like a junior member of your team that you would want to supervise. Usually we would recommend that teams using this in a jurisdiction have someone qualified in that jurisdiction to review the outputs. That's the key piece there, but it does work across jurisdictions.

One thing we have seen is that if a company has teams across Europe, as well as in the US and UK, they might have separate guidelines. German data protection regulations are very different from American data protection regulations, for example. The playbooks and knowledge bases Wilson runs off can be tailored specifically to each team's jurisdiction.

Case Law and Legal Research

Rod: You mentioned the US as one of your markets, and in the US so much is decided through court decisions. Very often these can be obscure decisions that happened somewhere many years ago. There are companies like LexisNexis and other legal databases where people can find what was decided decades ago. Can Wilson help there, either by interfacing with LexisNexis or by having this knowledge base to tell me what was decided in the past?

Gus Neate: Often we'd see people have crib sheets of what they're using as their key cases, and most teams would drive off those if they're a large organization with that sort of information internally. In terms of the product roadmap for legal research that I mentioned earlier, that would involve more of picking up these cases and interfacing with companies like LexisNexis or Thomson Reuters. That's further down the product roadmap for now.

Handling Ambiguity in Legal Advice

Rod: Also on the US market, there are situations where lawyers will provide advice, but it's never really clear-cut given that it might be litigated through a court decision. They might say there is option A and option B, but the decision-maker has to decide, and there isn't necessarily a clear recommendation. Can Wilson somehow advise here, or how can those who are making legal decisions but aren't in the legal space be supported?

Gus Neate: That's really interesting. I would say that Wilson's outputs would be quite similar to what you'd expect to see from lawyers, and we don't necessarily try to change that. Part of the reason lawyers give these options is that they want to set out the fact pattern, the risks, and the choices. It's often on the commercial teams, sales teams, marketing teams, or product teams to make those decisions.

Our tools provide options, setting out what you should be thinking about commercially and what the legal risks are. The choice often does remain with the business users.

Career Advice for Future Lawyers

Rod: I want to spend some time on the other side of the business—the lawyers themselves who are adopting the technology. In our audience, we often have people who are starting in their careers or in the early stages. Given that you've been a lawyer and are now providing tools for legal teams, what do you suggest for those who are thinking "I want to be a lawyer" or "I'm about to graduate" in the year 2025? With all this AI happening and automating so many processes, what should they be doing to future-proof their careers?

Gus Neate: I think the law remains a fantastic and extremely promising career path. It's an extremely important part of society, and the rigorous training you get as a lawyer really helps you to structure your thinking and understand businesses very well. I really enjoyed doing that at Clifford Chance, and I would recommend it to people. It was a fantastic experience.

I think in the modern environment, the key difference is that I would definitely stay close to AI tools and practice using them. They're changing the profession, and we're seeing people adopting these tools. Being really close to them is very valued within organizations. If you jump into an in-house legal team or a law firm, being the person who really understands how these tools can be effective for the team and can drive adoption—that's extremely valuable as well.

Rod: So future lawyers should not fear AI but rather embrace it.

Gus Neate: Yes, exactly right.

Final Thoughts and Key Takeaways

Rod: We've covered quite a lot of ground, but is there anything you'd like to share with the audience that you find important for them to know about Wilson?

Gus Neate: There are probably a couple of things. One thing we've seen from historic tools is that there can be an assumption that using an AI tool requires a huge amount of training data or a huge amount of implementation. But what we're starting to see is that newer tools, because they can work across unstructured data, are able to be implemented and used a lot more quickly.

We often see customers saying, "It took me 12-18 months to implement this contract lifecycle management system." What's interesting is that those processes can be massively shortened by the fact that Wilson can be trained across unstructured data and help you with structuring it. That's one really interesting development—how much easier it is to get started with these tools than it used to be.

The other piece I'd love to mention is that sometimes people think about this as "AI versus humans," and it's quite binary. The way we view this is that AI is really going to augment how people are working and support people in doing their jobs and enjoying them more. Many of the tasks that people don't enjoy in their job—searching for a particular point, reviewing it repetitively, or answering the same question over and over again—those items will be picked up by AI. This means that people are doing the strategic work of looking around corners. It's really AI with humans hand in hand—it's about augmentation.

Getting Started with Wilson

Rod: I'm sure that after this conversation, there will be some who say, "I want to get started with Wilson. I want to give it a try." How can they do that, and what does the integration process look like?

Gus Neate: They can reach out through our website www.getwilson.ai. The integration process is actually super simple—it's a one-click integration to pull all your information in from Google Drive, for example. You can select the documents after clicking that. The same thing applies for Notion or OneDrive. You can just drag and upload your documents.

All this takes around five to ten minutes to get set up. Then you can deploy Wilson with one click into a Slack channel or start asking questions through the web app. It's super easy to onboard.

Rod: So for anyone who wants to get started, go to getwilson.ai. But if someone has any questions or wants to get in touch, where can they find you?

Gus Neate: They can find me on LinkedIn, or my email is gus@getwilson.ai. I'm super happy to take any questions. I'm based between London and San Francisco, so I spend time in both places. Our team is in London, and I go to San Francisco every six weeks or so. I'd love to meet up in person for a coffee to discuss more, or I'm very happy to speak online. People can book demos through our website as well.

Closing Remarks

Rod: Great! We've covered so much about what Wilson is doing, the current situation with legal teams, what early career lawyers should be doing to prepare for the future of AI, and where Wilson is headed. I'm sure many in the audience will be interested in reaching out, either to understand what's happening in the legal space or to start adopting Wilson.

Thank you so much for coming here today, Gus. It has been a fantastic discussion. And for everyone else, remember, leave us your comments. Let us know what you think. Are you already adopting AI for legal? Are you afraid of AI for legal? What are your thoughts? And don't forget to join our newsletter on chrisrodmax.com. Thanks so much, Gus.

Gus Neate: Thanks very much, Rod. Really appreciate it. Cheers.

Your Hosts

Christine Wang: https://www.linkedin.com/in/christinewang0/

Rod Rivera: https://www.linkedin.com/in/rodriveracom/

Maxson J.Y. Tee: https://www.linkedin.com/in/maxsontjy/

Social:

X: https://x.com/chrisrodmax

Instagram: https://instagram.com/chrisrodmax

LinkedIn: https://linkedin.com/company/chrisrodmax

Subscribe: https://chrisrodmax.com

Thorsten Kranz, head of AI solutions at DHL Group, joins Chris Rod and Max in this episode. They explore the challenges of AI adoption, the importance of data quality, and the impact of AI on various business functions.

The conversation also delves into investment trends in AI, the decision-making process between building or buying AI solutions, and the potential disruption of jobs across multiple industries due to AI advancements.

In this conversation, Thorsten Kranz discusses the evolving landscape of AI, its applications across various industries, and the implications for the job market. He emphasizes the importance of adaptability and continuous learning in the workforce, as well as the need for managers to lead by example in adopting AI technologies.

The discussion also touches on the challenges posed by AI in terms of skill gaps and the potential for societal divides based on AI literacy.

Chapters

• 00:00 Introduction to AI in Enterprises

• 02:51 AI Adoption and Organizational Readiness

• 06:34 Data Quality Challenges

• 10:10 Interactive Data Analysis and User Engagement

• 17:17 AI’s Impact on Business Functions

• 21:09 Investment Trends in AI

• 25:38 Building vs. Buying AI Solutions

• 30:39 The Future of Jobs in the Age of AI

• 31:58 Exploring AI Use Cases Across Industries

• 35:27 The Future of AI and Job Market Dynamics

• 41:13 Hiring for the AI Era: Skills and Mindset

• 47:30 Practical Applications of AI in Management

• 55:45 Final Thoughts on AI’s Rapid Evolution

Takeaways

AI adoption requires robust organizational foundations before meaningful benefits can be realized: While many companies are rushing to implement AI solutions post-ChatGPT, Thorsten emphasizes that success depends on having proper data infrastructure, consistent management practices, and organizational readiness in place first, noting that without this groundwork, companies won't see the expected returns on their AI investments.

The future workforce will likely split between AI-dependent workers and AI-proficient specialists: Thorsten predicts a significant bifurcation in the job market where most workers will heavily rely on AI tools without deep understanding, while a minority who truly comprehend and can go beyond AI's capabilities will command premium compensation, potentially leading to wider economic disparities that might necessitate solutions like universal basic income.

Technical leadership in AI requires hands-on engagement and continuous learning: Drawing from his personal experience, Thorsten emphasizes that effective AI leadership isn't just about directing others - it requires leaders to actively use and understand AI tools themselves, maintain technical proficiency, and constantly experiment with new capabilities, suggesting that theoretical knowledge alone isn't sufficient for guiding AI transformation.

YouTube Episode

Spotify Podcast

Transcript

Chris: Hello everyone, welcome to another episode of the Chris Rod Max show. Every week, we discuss the latest news trends and dynamics within the AI space with distinguished guests. Today, I'm delighted to be joined not only by Rod and Max but also by Thorsten Kranz, Head of AI Solutions at DHL Group. Thorsten, welcome to the show.

Thorsten Kranz: Hi, and thank you for having me. I'm looking forward to our discussion.

Chris: Before we dive into today's fascinating topics, could you share a brief introduction about yourself and your background?

Thorsten Kranz: Of course. At DHL, I manage all aspects of AI, which has become increasingly intense over the past two to three years. My journey in AI began with physics studies, followed by neuroscience research around 2007, where I applied artificial intelligence to understand natural intelligence. Later, I joined a consultancy as their first data scientist, building machine learning solutions across various industries. The team grew rapidly – from just me to 20 people in two years, eventually expanding to a department of 50.

I then decided to move from consulting to join a company's long-term journey, which led me to DHL. While I'm now in management, I'm still a technical person at heart who loves to code and work with data. I've found a way to combine both passions in my current role.

AI Implementation and ROI in Enterprise

Chris: That's fascinating. You've truly been a trendsetter as one of the first data scientists in your company. Today, we'll explore three main topics: your perspective on AI in enterprise, careers and jobs in the AI era, and your personal experience with AI implementation. Let's start with a recent Lenovo report about AI's return on investment.

The statistics show that 26% of adopters say AI projects have exceeded expectations, while about 70% report met expectations. However, many organizations haven't adopted AI solutions yet – 25% are still in pilot phases. A major obstacle seems to be low organizational readiness in terms of data, processes, and IT infrastructure. Thorsten, what's your reaction to these findings?

Thorsten Kranz: These findings aren't surprising. Since ChatGPT's release, we've seen many unrealistic expectations. Some companies are just starting their AI journey, while others may have neglected fundamental work on data consistency, proper management, and organizational structure.

Without proper groundwork, you won't see benefits. At DHL, like many companies, we've been working in this field long before ChatGPT. You need to prepare your organization and IT infrastructure, including MLOps capabilities to easily transfer proofs of concept into production scenarios. This includes monitoring, logging, and support infrastructure.

Even with the advent of large language models, core requirements haven't changed. You still need to integrate AI into your processes and ensure people understand and use the tools effectively. If you're just starting now, there's still significant work ahead.

Data Quality and Organizational Challenges

Chris: That makes sense – you need strong foundations, particularly data quality. I remember a meme about people thinking ChatGPT is magic, but with poor data, you'll get poor results. What makes it so difficult for organizations to maintain good data quality?

Thorsten Kranz: There's no single answer – organizations vary greatly. An e-commerce company with one business model might have relatively simple data needs: website usage and stock information. But historically established companies, especially in Europe, face different challenges with thousands of legacy systems and various potentially relevant datasets.

You need clear prioritization, focusing first on the most valuable data likely to provide ROI. In my experience, initiatives solely focused on improving data quality or governance typically fail. Data quality improves through data usage – only when poor data impacts results do organizations make improvements.

It's best to work iteratively: improve data quality, use the data, identify issues, return to improve the data, and modify the processes that generate the data. This creates an upward spiral of continuous improvement rather than a long-planned program.

Legacy Systems and Modern AI Integration

Rod: Regarding legacy systems, how do organizations make diverse, possibly outdated data sources available to users, especially for AI applications?

Thorsten Kranz: It's an ongoing challenge. Years ago, the reporting world evolved with dashboards for everything – now people need dashboards just to navigate their dashboards. The promise of accessible data has become a maze.

We're seeing a shift toward interactive data analysis for non-technical users, especially with large language models and chat-style interfaces. This allows spontaneous access to insights like employee demographics or gender ratios at different management levels. Rather than creating another dashboard, we're enabling direct data access and insights.

User Adoption and Training

Rod: Some users struggle with formulating thoughts in text prompts. How do you help users interact with these open systems effectively?

Thorsten Kranz: We tackle this from multiple angles. We've invested heavily in training and enablement. Before ChatGPT, we rolled out extensive data science training, including full-day management sessions that thousands of managers have completed. Tens of thousands of employees have completed e-learning courses on data concepts and quality.

With ChatGPT and newer tools, we've established additional training formats. Beyond awareness sessions, we provide dedicated training on understanding large language models' capabilities and limitations. We teach prompting techniques, though newer models like Claude-3 and DeepSeek R1 often understand intent better than earlier versions.

ROI and Business Impact

Chris: Speaking of return on investment, which business functions have shown the highest potential for AI-driven returns?

Thorsten Kranz: Customer service is an obvious area, particularly verbal and written communication. The natural feel of advanced voice models from ChatGPT, Gemini, and others is impressive. Most organizations haven't fully leveraged this technology yet.

However, commercial considerations are important. Some vendor rates are below European labor costs but above offshore rates. We need to ensure AI gains benefit us rather than creating vendor dependencies that consume the ROI.

Beyond customer service, AI shows promise in scalable, repetitive, text-based tasks. But everything must be digital first – many companies still struggle with end-to-end digital processes and documentation. Only with digital historical examples can AI learn and automate effectively.

AI Investment and Infrastructure

Chris: Regarding AI investment, major US tech companies are pouring money into AI – about 63% of their spending. Industry-wide, nearly 20% of tech budgets in 2025 will go to generative AI. What are these investments focusing on?

Thorsten Kranz: The biggest investment is in GPUs. For example, xAI reportedly has 100,000 H100 GPUs and may double that soon – investments on par with major German company valuations. This explains NVIDIA's dramatic share price growth.

Beyond hardware, talent is crucial. We're seeing a talent war with incredible salaries for experts. Infrastructure is another challenge – companies are even investing in dedicated nuclear plants for AI computation centers.

The most investment goes into training foundational models, which increasingly become low-cost or free to use. Value creation may shift to the application layer – applying these models to specific domains like IT ticket resolution. The question is how much of the savings will benefit end-users versus vendors and model creators.

Build vs. Buy Decisions

Max: How do you decide between building AI solutions internally versus using vendors?

Thorsten Kranz: For smaller companies, I strongly recommend against building too much internally. Creating a prototype is relatively easy, but scaling and maintaining it requires substantial infrastructure. Get help from partners – they don't have to be big tech companies; smaller consultancies can work well.

At DHL's scale, we have large teams and experienced colleagues, but our decision process is complex. For standard tasks, we expect multiple vendor solutions, giving us negotiating power. More importantly, we focus on logistics-specific niches where vendor solutions don't exist, like customs clearance automation.

Future of Jobs and AI Impact

Chris: There's an interesting A16Z article listing top 50 jobs that could be AI-automated, spanning architects, financial analysts, dental services, and more. What's your perspective on this widespread potential disruption?

Thorsten Kranz: I believe any limitation of AI will vanish over time. Everything humans can do, AI will eventually be able to do. The best response is to actively engage and shape developments. While I support some regulation, the European approach may limit innovation, especially for smaller companies.

Advising my children (12 and 15) about careers, jobs heavily reliant on personal relationships may be harder to replace. The key is staying curious, open-minded, and accepting that careers will require continuous adaptation.

AI Literacy and Social Impact

Rod: Studies suggest we're becoming "AI illiterate" by relying on answers without understanding. Your thoughts?

Thorsten Kranz: It's a real challenge requiring awareness and reflection. For some areas, like calculator use, we've accepted automated results without verification. I expect a bifurcation: most people will rely heavily on AI, while a minority will maintain deeper understanding.

This creates societal challenges – why pay high salaries to those who simply rely on AI? The minority who go beyond AI's capabilities will command premium compensation. This could widen economic gaps, making universal basic income more relevant as value creation concentrates among a small workforce segment.

Personal AI Usage and Leadership

Chris: How do you personally use AI in your daily work?

Thorsten Kranz: Managers must lead by example – I've seen many urge teams to use AI while barely understanding it themselves. I maintain "street credibility" by actively using various AI tools and subscriptions, comparing capabilities like voice modes between ChatGPT and Gemini.

My most important use is organizing thoughts and converting abstract ideas into actionable plans. Using voice mode or transcription, I can dump thoughts into AI systems, which help structure them, add external knowledge, and create presentation outlines. This has transformed my approach to new concepts – what might have taken a day now takes 30 minutes.

I never start texts from scratch anymore, and I regularly experiment with AI development to understand implementation challenges. There's hardly anything I do without AI these days.

Balancing Technical and Management Roles

Rod: How do you balance hands-on technical work with management responsibilities?

Thorsten Kranz: It's challenging – my coding often happens late at night due to "revenge bedtime procrastination." This is only possible because it's my passion. My day prioritizes work, then family, and finally technical exploration when energy permits.

AI and the Next Generation

Max: How are your children using AI?

Thorsten Kranz: They use AI more thoughtfully than most their age, given my guidance. They use it for schoolwork but not to complete assignments – rather as a learning aid and orientation tool. It's about using AI as a buddy while maintaining independent thinking.

Closing Thoughts

Chris: Any final thoughts on AI and its future?

Thorsten Kranz: Three key points:

1. AI's power is no longer on the horizon – it's happening now. Showing today's models to someone from three years ago would seem like AGI.

2. Don't be fooled by specific failures. AI's capabilities are unevenly distributed – it may fail at simple tasks while surpassing PhDs in complex domains.

3. The future is uncertain – we don't know where value creation will concentrate or what will be possible. We must stay adaptive personally and organizationally.

Chris: Thank you for these insights, Thorsten. The key takeaway seems to be "lead in rather than ignore and stay nimble." How can our audience reach you?

Thorsten Kranz: You can find me on LinkedIn and other typical channels. Happy to connect.

Chris: Thank you everyone. If you enjoyed this episode, please subscribe to the chrisrodmax.com newsletter and channels.

Your Hosts

Christine Wang: https://www.linkedin.com/in/christinewang0/ 

Rod Rivera: https://www.linkedin.com/in/rodriveracom/ 

Maxson J.Y. Tee: https://www.linkedin.com/in/maxsontjy/

Social: X: https://x.com/chrisrodmax 

Instagram: https://instagram.com/chrisrodmax 

LinkedIn: https://linkedin.com/company/chrisrodmax 

Subscribe: https://chrisrodmax.com 

Tags: #chrisrodmax #ai #technews

In this episode, Ali Raza, CEO and co-founder of Bevel, discusses the challenges of working with legacy code and how Bevel aims to provide solutions through AI-powered tools. The conversation covers the definition of legacy code, the importance of understanding business logic, and the vision behind Bevel as a company.

Ali shares insights from potential users, the approach Bevel takes to code understanding, and the competitive landscape in the software development industry. The episode also includes a demo of Bevel’s tool, showcasing its capabilities in navigating and modernizing legacy code.

In this conversation, Ali from Bevel discusses the challenges of legacy code in software development, the urgency for companies to modernize their systems, and the role of AI in facilitating this transition.

He emphasizes the importance of understanding legacy systems, the need for a fulfilling workplace culture, and the strategies for engaging with the developer community. The discussion also covers the implementation of Bevel’s solutions, data privacy concerns, and the evolving pricing strategies for their product.

Chapters

• 00:00 Introduction to Bevel and Legacy Code

• 05:59 Insights from Potential Users

• 08:58 Bevel’s Approach to Code Understanding

• 11:54 Company and Product Vision

• 14:52 The Future of Software Development

• 18:06 Demo of Bevel’s Tool

• 20:56 Navigating Legacy Code with Bevel

• 24:05 Competition and Market Positioning

• 26:50 Real-World Applications and Challenges

• 30:43 The Tipping Point for Legacy Systems

• 32:41 AI’s Role in Modernizing Legacy Code

• 34:36 Agentic Software Development

• 36:24 Implementing Bevel in Legacy Environments

• 37:30 Data Privacy and User Experience

• 40:11 Language-Specific Solutions for Legacy Code

• 42:42 Building a Fulfilling Workplace Culture

• 45:07 Navigating Early-Stage Startup Challenges

• 46:14 Community Engagement and Product Adoption

• 47:58 Pricing Strategies for Developer Tools

• 49:37 Final Thoughts and Call to Action

Takeaways

Legacy Code is a Massive but Often Overlooked Challenge

Legacy code, defined as code written without modern documentation and testing practices, is still the backbone of critical systems in industries like banking, insurance, and energy. Many enterprises struggle with maintaining and modernizing these systems, with trillions of dollars spent annually just on upkeep. The risk of key personnel retiring without adequate documentation makes the problem even more pressing.  

Understanding, Not Just Rewriting, is the Key to Modernization  

Bevel's approach emphasizes that the true bottleneck in legacy code isn't the code itself but the loss of business logic and application knowledge over time. Instead of simply transpiling old code into a modern language—often a flawed approach—Bevel focuses on abstracting and understanding the business logic, allowing organizations to maintain, modernize, or migrate their systems intelligently.  

AI-Powered Code Comprehension is the Future of Software Maintenance

With AI-driven code understanding and visualization, Bevel provides deterministic insights that help developers navigate complex, undocumented codebases. Their product enables enterprises to proactively document, maintain, and modernize their systems in a secure, privacy-first manner, ensuring long-term stability and efficiency. As AI-generated code becomes more prevalent, tools like Bevel will be essential for making sense of both legacy and AI-generated codebases.

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Company Overview

Rod: Welcome to another episode of the Chris Rod Max show. This week we're interviewing Ali Raza, CEO and co-founder of Bevel. Hi Ali, welcome. Great that you're here with us today.

Ali: Hi Rod, really nice to be here. Thanks for the invite and I'm looking forward to the conversation as well.

Rod: So before we get started, let's start by defining Bevel in one sentence. How will you define it?

Ali: Bevel is an AI-powered tool that helps developers that work with legacy code.

Understanding Legacy Code

Rod: So legacy code, that doesn't at first glance sound like the most exciting topic for a startup. How did you get there? Let's talk a little bit about where you were and how you're now. What's your background, your path?

Ali: Yeah, sounds good. Maybe even before that, I should define legacy code because unless you really know what it is, it's very hard to understand. In very technical terms, legacy code is code that doesn't have test coverage, but that doesn't really mean anything. For the layman, legacy code is basically code that was written in a time where code hygiene was not a thing. Imagine you work in a company and you have a code base that has millions and millions of lines of code.

You have no idea how things are written, how they are connected, because everyone who wrote it has left the company or retired, and there is no documentation. So it's very hard to work with such complex code bases. That's what we mean when we say legacy code.

The Birth of Bevel

And you're exactly right - it's a very niche, very specific problem. And we actually came across it very accidentally. The way Bevel started was it was very team first. I come from CDTM - Rod, you're also an alumni of that. My other two co-founders are also from the same class there. We basically started very team first. We said, we have very complementary skills, we want to work together, and we want to start a company someday. But we had no clue what it was going to be about.

What we did was we did hackathons in Munich. Every other weekend, we'd get together to do a hackathon. And at one of those hackathons, my co-founders actually built a code understanding tool. It just parses the code and visualizes all the dependencies. Then we got invited by an insurance company here in Munich. They asked us to present the idea to the team and they said, "You know what, we have this very large monolithic code base and it would be super cool if we could use something to bring understanding to it."

And we said, okay, sounds cool, but we don't want to just build for you, we want to build something bigger. We talked to more people, realized it's not just them, it's a huge problem in many different industries. So that's how Bevel was born.

The Scope of Legacy Code

Rod: Very good that you define what is legacy code because not everyone knows that. Whenever I think of legacy code, I think maybe somewhere in some bank, there is something that was written back in the 60s. And now it's gone completely, not only has the person retired, but there is no more any way to just understand what's going on. But yet this is powering maybe some important system and nobody can just go and say, hey, I'll shut down this button and then maybe replace it because it's not even clear what's the logic.

Ali: 100%, 100%. Every time you swipe an ATM card, think 95% of the time there is some legacy code in the background that's being executed. So our modern world is kind of running on the shoulders of legacy code in a way.

Early Development and Customer Discovery

Rod: And so you're also mentioning here, so CDTM for those who do not know is a program based in Munich. It's a study program with a focus on entrepreneurship. And you basically met your co-founder at this hackathon. So you had this tool that can visualize code. I imagine that then, this insurer looked at it and said, hey, it looks exciting. Were you already expecting what you would see, did you already know at that time that this could help with legacy code, or did that come once you were in there and saw what's going on at that company?

Ali: I think we were very much, because it's a young team, we just thought it's a cool tool that would help with any code base. So even a smaller, not legacy code base, can use some kind of understanding. That was the kind of framing. But then we saw this code base, and then we talked to more people. We did more research. And we realized, this can actually be used for a huge problem that is very pertinent and very persistent across many different industries.

In the beginning, we were quite naive about it. We were like, no, it's just a cool thing we built over the weekend, and we can use it for our own code base. And then I think it was more serendipity that we came across the problem itself.

Market Size and Impact

Rod: When you try to quantify it, when you try to put it in numbers, how would you define how relevant is this problem?

Ali: Very, very big numbers. Firstly, if you really look at legacy code itself, there are stats that out of the top hundred banks, 92 of them still have some legacy code on COBOL in the mainframe somewhere. Even all the really cool banks with really cool front-ends - everything that is user-facing - but a lot of them still use mainframes for batch processing.

Almost 80 to 90 percent of all Fortune 500 companies have the same issue. They're still running software written a long time ago, and there was this number - I think it was like $1.73 trillion that every year is spent on just maintaining legacy code. So it's a huge and very persistent problem across industries. Banks is just one example. You can think of banks, insurance companies, automotive, energy. So all of the tried and tested industries, they all have the same problem.

Customer Insights

Max: Just a quick question, just to jump in there. Based on your conversation with some of the potential users of Bevel, what was the most interesting insight that you got from them? Because I mean, you built the tool trying to understand legacy tools, trying to understand code and legacy code. What was the most interesting insight you got from some of your potential users?

Ali: Yeah, so I think the most interesting inflection point for us was that it's actually not about the code. We started with the code itself. We said, code is hard to... COBOL, for example, is a... For anyone who doesn't know, it's a language that's mostly used in banks. And we said, okay, it's hard to understand and code understanding is a huge bottleneck. But then we realized that's actually not the problem.

Code is not the problem. Neither is it the solution. The code is trying to do something. And this is what Rod already mentioned - it's the business logic. The code is trying to accomplish something. But in many cases, that's lost. So people who know what the code was written for are not there anymore, and there is no documentation.

So it's actually this higher abstraction of understanding - what is the business logic, where is it implemented? And if I change something in the code, will I break some critical business logic? That's the more important or harder problem to solve. COBOL itself, the language, is super easy to understand. It was written kind of more for business people actually, so it's very descriptive. You can find developers - if you go on LinkedIn, you will find a lot of developers looking for a job for COBOL. So that's not the bottleneck. The bottleneck is understanding the application knowledge that is lost.

Product Functionality

Rod: So then therefore the idea is to try to create some sort of documentation for this code. You mentioned that part of the initial concept was to create these diagrams that will visualize the business logic for the code. So is it maybe a little more focused on helping get an overview of what it's doing from this perspective, or is it more focused on moving from this older legacy code base to where something is modern and sexy?

Ali: What you said is the former one, but I think by definition that helps the second thing that you mentioned. The first thing is what is the business logic? Where is it implemented? And that's firstly a UX problem as well. It's a technical problem because firstly, it's very hard to just get that from the code itself. You need a lot more sources. You need something like a data dictionary, for example, that most companies don't have. So you need to enrich and give more context to the tool itself to really understand what is the business logic and then where it's implemented.

The way currently we do it is we have this one table where all of the functional requirements, all of the business logic is stored. Then we have these diagrams that help you trace through the code where it's implemented. Then we have this code lens that's in the code base itself where you can just click and see what is the documentation for it. And the documentation is always up to date - if you edit it, it stays that way and it remembers that.

It's always up to date. But that's what we're at. And then from there, you can go two ways. You can say, I want to just maintain the code - so I understand the documentation, I can maintain it. The second one is you say, OK, I want to modernize it or migrate it. Because for migration, usually what people try to do was they try to transpile the code. They said, we have this COBOL, let's translate it to Java, which doesn't really work because it's two very different languages, two very different frameworks of doing things.

And now you have Java, but it still has the same problems as the initial code. So what modernization really means is you understand or abstract the business logic, and then you re-implement it. And that's the thing, that's the missing link. And that's where we provide the context, not just for people, but also for other tools. So you can use Cursor or GitHub Copilot to rewrite the code, and we give the context to those large language model tools as well, so they can use us and then re-implement.

To summarize, I said a lot of things - we focus on understanding, but understanding can be used for maintenance as well as migration and modernization.

Company Vision

Max: Great, think that's very clear. And in terms of, I guess, the vision for the company, is it to change all 93% of all banks' payment systems because they're horrible? Just saying. So I'd love to understand the vision as well.

Ali: There's actually two visions we have that are slightly different. So one is the company vision, the other one is the product vision. And I'll start with the company vision. The company vision is actually not about legacy code. So people always find this very surprising because if you go to our website, our vision is we want to make Bevel the most fulfilling workplace to work at.

And the idea behind that is traditionally, companies always wanted to maximize shareholder value. So they said, let's do everything that we can to maximize shareholder value and increase profits. Then there was a second wave of companies like Amazon that came around and they said, it's customer first. We can sacrifice short-term profits for long-term gains because if we take care of the customer, the profits will follow.

And we took one more step and we said, let's take care of our team. If you have a very fulfilling workplace - and by fulfilling we don't mean that every day everyone is very happy and cheery - it's more that everyone feels fulfilled to show up to work every day and they feel like they're doing impactful work. If we have that in the team, we can take care of the customers and the profits will follow.

So that's the company vision. For the product vision, Bevel is a product that we are building and the product vision is we want to take on all legacy code in the world. There is a lot of that, but in the long-term future - like really long-term future - we want to transition or actually leverage what we already have into code understanding.

Because the way software development is going right now, a lot of code generation would be commoditized, at least in our view. So there would be a lot of code written by AI. You would have fewer software developers who would work on a higher level of abstraction to manage and maintain this code. And because a lot of this code will not be written by them, they will need understanding.

So that's our long-term vision. Firstly, help legacy and then use the same technology, the same product to go into this - how can we help developers understand code much better that was not written by them.

Future of Development

Rod: Interesting. Before we go to the demo Ali, I'm wondering, so you're basically saying there will be less and less developers. There is also like counter argument that people are saying hey because now it's getting so easy to write code through AI, pretty much everyone will become a coder. So just like we are writing now an email, people will write programs and therefore it will be that all of a sudden everyone is a developer. Do you think that this will also happen or when you say there will be less developers, you mean professional developers? How do you think this will develop?

Ali: Yeah, I think let me correct myself a bit. What I mean is less developers per application. So it will take less developers to write, maintain, and run an application. Maybe there will be more applications and hence more developers overall. I think it's very hard to make that call right now. But for one application or one company, there will be less developers because writing code and generating code, all of those things will be already much easier.

So that's why we think code understanding is the thing, because a lot of this code will not be written by them, and they will need help understanding and maintaining that code. Yeah, I think to close the loop, I think it's very hard to make a call right now, which way it will go. I think maybe everyone will be a developer, but probably writing an application and running and maintaining it will be so much easier. We'll just require a different level of thinking and abstraction. The role of a software developer will just evolve in a way.

Market Opportunity

Max: Yeah, I think just to add on to what you were saying, I think the thing about AI is we don't really know what will happen every three months. Some deep sick will come again. Maybe the next one is sick deep. Who knows? The funny thing is, I guess, you know, there is an opportunity on the legacy code base, right? If a lot of banks are still running or a lot of larger companies really still running on legacy code base, I think there's a huge opportunity.

Even from there, if you were to just try to modernize some of those, because they are critical infrastructure in which you can't go down. I think, I don't know if you saw, I think over the weekend, Rod, in the UK, Barclays obviously went down. Yeah, so I can see how a lot of the banks would jump on this, right? Just try to help themselves to be more resilient, not just for themselves, but for customers and also regulators asking for it. So it's quite interesting.

Product Demo

Rod: And I'm sure that for many in the audience this is definitely a massive problem, but I feel that from an audience perspective, especially those who don't work at a bank or aren't reading code, they might not really realize how relevant legacy code is. So I wonder if you can show us how it looks in practice. I don't know if you have some legacy code database or that we can see how we can go from this legacy situation, this chaotic code base towards something that is easier to understand.

Ali: I wish I could show you the ones that we're working with right now. So there is one which is five million lines of code. And I remember we got on the meeting with the company and we were talking and I asked them, can we use some documentation that you already have to benchmark our performance? The documentation we generate and the documentation you already have to compare them. And I remember they started laughing and they said, there is no documentation.

So there's 5 million lines of COBOL with no documentation. But most of these code bases are proprietary and closed source. So it's very hard to find them unless you're working with those companies. I do have a small pre-recorded demo video that I can pull up and show around if that's interesting for you.

Rod: Yes, it will be very interesting just to see because I also have many questions surrounding the development of product itself, given that you're working with code that was written decades ago and it's not, say, like the popular programming languages of today that there are so many resources online available. So if you have any demo walkthrough, please go ahead Ali.

Ali: I have a very small demo recorded by my co-founder Juan. He's our genius on the product presentation side as well. So I'm going to give the mic to him. Let me know if you see my screen.

Rod: Yes, you can see the screen.

[Demo presentation occurs]

Product Features and Technology

Rod: That's really good to see and have an impression on how it all works in practice. And earlier in the video, we're looking at basically an interaction diagram. So the idea is that here, the user can understand the logic behind, or does it just help document it better? So how do these visualizations play a role, especially because you mentioned earlier that you started pretty much as a tool to visualize code?

Ali: Yeah, so the diagram you saw was a sequence diagram, so it shows how data flows through the code base itself. And there we have actually a UX problem to solve, which we are solving, which is even if you visualize everything in a code base that has 5 million lines of code and I don't know how many functions and how many classes, it's very hard to give the right context to developers because you just get lost.

The goal is you have this documentation, but we still want you to be able to navigate the code in a way, to understand how things are connected to each other. So that's where the diagrams come in, because there's more visual way of understanding how things are connected, how data flows, and all of those things. And then you only see what's very relevant to you. And so one of the fronts we're working on is how can we make this much more easy to navigate, much more relevant for exactly what you want to do.

So if you have some very specific task in mind, you only want to see certain things. You don't want to see anything else. And how do we show you the right thing? So that's a UX problem we have. I think one more thing to mention here, which differentiates us from other tools - so there's some other tools just for code understanding, they just have this one feature that we have, diagram generation. They are not deterministic. So what they do is they take the code base, they try to feed it into an LLM and ask it to generate a diagram, this is not deterministic. So if you use a different model, if you use it twice, it might give you different diagrams.

What we are showing you is 100% deterministic. Every single time you see the same thing. And that's because we are leveraging tried and tested technologies. So we're using static code analysis. We're using a language server from VS Code, which means that all of the diagrams, everything there will always be the same.

Competition and Market Position

Max: Ooh, I think just while you're talking about that, I was thinking about competition-wise, you did mention VS Code, and I know quite a lot of large enterprises have some sort of static code analysis. A lot of times, I think they do it for security reasons, but I'd love to understand how do you think about competitors in this space. Sometimes it could also be a moot point because things move so quickly. But just love to hear how do you think about this.

Ali: Yeah, so firstly, I think our biggest competitor at least was just companies doing nothing about this. But I think now we are at this inflection point where a lot of people are just leaving the company - the developers in these companies are just retiring in the next couple of years. So they have to do something about it.

The second one is actually consultants. So there's a lot of IT consultants that actually do this, right? So they do modernization, migration, also sometimes maintenance here and there. But they're very expensive, firstly. It takes a lot of time for them to understand it and then even migrate it. So it takes at least a couple of years, five to 10 years. And by that time, the code is legacy again, because your own developers don't really understand it anymore.

So that also doesn't solve the problem. It just puts a patch on it. And we are also leveraging them - so we see them as our partners in a way. We basically use the consultants and we use them as a distribution channel because we also help them understand the code much faster.

And then we have all of these tools that comes to everyone's mind - GitHub Copilot, Cursor, everything, right? But they're very good at code generation and code understanding. But everything before that, like understanding of the business logic and all of those things is very hard to do and it's a very different problem. So we are very complementary to them.

If we can abstract all of these things out, and we can help companies build these things, this is the missing link between then using Cursor or GitHub Copilot or something else to generate new code. Then we have the static code analysis tools, like Sonar, for example. But what they do mostly is metrics or code quality, all of those things. And they're not really focused on understanding of the business logic.

Then of course we also have code understanding tools. There's also some YC company recently. They're mostly doing non-deterministic generation of diagrams from code to help you understand it. Because deterministically drawing these diagrams is not a trivial technical problem to solve, especially for large code bases.

That being said, I think at a startup we have the mindset that at least 20 teams are working exactly on what we're working on while we're working on it because the space is so interesting and it's ripe for innovation. For sure there's a lot of people working on the space as well. And one of our strategies is also to be faster than them. And I think distribution is the biggest bottleneck. So we want to get to the enterprises, get to the customers as soon as we can with a really good product that helps them.

Real-World Examples and Challenges

Max: Yeah, I think one good thing you have going for you is that today a lot of all these larger organizations are not doing anything with it in a way. So helping them to understand those legacy code bases, I can see the value of this. I mean, I have a story to tell, right? Like on this, I know there is a very huge critical system that's running a lot of the financial services in the world. It's actually run on this old language called Toptop.

So what happened is when they bought it, they wanted to just translate it. They literally just used a translator tool and turned it into Java. And let's just say it's not pretty. Just plenty of problems after problems after problems. I can see how your tool can just drop inside. Look, this is how it works. And not to mention, people tend to write their own code on top of the code. I was like, god, what are we doing here? So as I can see, there's a huge, huge opportunity. Exciting.

Ali: Wow, yeah, 100%. And I think people also underestimate, when we say legacy, people underestimate how legacy we mean. And that's why a lot of these questions come around using other tools like Copilot for this. I'll give you an example, actually. So yesterday we were talking to this company that's basically migrating a core banking code base from COBOL to Java as well. And they have tried all of these transpilers, right? And they don't work because they want to keep the same way of doing things.

They just want to understand what it did and now re-implement it so it interacts with the database in a different way, but implement it in a different way, right? And they also tried, I think, something from IBM, Watson X is called, that also does something like code refactoring automatic. Also did not help them. So when we say legacy, it means really, really legacy.

And one of the problems they have is the code base has many different symbols that it's very hard to define them. And usually you have a dictionary - so you need to have a data dictionary that defines, OK, if this symbol is used, this is what it means. It's not standardized. It probably is different for every company as well. And all of these tools don't really have this context.

So the way we are doing it is we're also helping them create these data dictionaries on the side and then use it for our tool to help them understand it. And this is, let's say, another moat that we have, that when we say legacy, we actually really mean legacy.

Historical Context and Timing

Rod: The problem of legacy code has been around now for decades. I remember, like in the late 90s, there was a discussion of Y2K, all this old code will not be able to adjust the date for the new millennium. So as a problem, I would say it has been known for quite some time already. And I wonder here, when you speak with companies, what's their succession plan or the solution to this?

So you were mentioning that one of these is, for example, consultants, maybe like offshoring. However, do companies have an idea in mind what to do with it? Because for example, Max was mentioning so many important systems are still run in programming languages that are by now many decades old. Even if we think something like Java, Java is from the early 90s, but by now it's starting to show its age. So therefore I wonder here, what are companies doing? What are they telling you?

Ali: Yes, you're exactly right. So the problem of legacy is as legacy as the problem itself. So it's quite old. And so the question comes from there, why now? Why is now the time to act? And there's a couple of answers to that.

The first one is, finally the problem has hit the tipping point. And an example for a bank would be that most of the people working, that know the application, are about to retire. So I was talking to the CTO of a bank, of a large German bank, and he told me there's a core module they have - only two people understand it. The only two people have understanding what's going on in there. Both of them are going to retire in the next one or two years. And they have no plan for that whatsoever.

And this is not just them. It's everywhere. There's examples of companies bringing people back from retirement, paying COBOL developers who understand the application, paying them a lot of money to just come in as a contractor and try to still run those systems.

Secondly, we finally have the technologies. With developments in AI, we now finally have the technology. So with static code analysis, you can only go so far. But now we finally have the technologies to come at the problem. So we have now the capability to retroactively document these systems, create some kind of understanding around it, which we didn't have before.

And I think thirdly also, banks and all of these companies face a lot of competitive pressure to be lean and to be competitive with other companies. And these software systems slow you down. It's very hard to add on top of that. It's very hard to extend functionality and all of those things. And all of these companies, they want to be agile and they want to be competitive with others.

And I think finally, it's like at the end of the day, even within companies, it's people, right? And one of the things that CIOs, for example, banks have on the top of their head is, what do we do about AI? I don't want to be the person who does nothing about it when everyone else is doing it. So there's also that around it. And all of these together, it paints a very compelling picture of why now is the time to act and do something about this.

Product Implementation and Data Privacy

Rod: Once a bank, for example, has understood, hey, we need to do something with this legacy code, and not only that, but also AI is this powerful tool that will help us - how do they go about implementing Bevel? So they say, hey, Ali, we want to have Bevel in our organization. Is this plug and play? Because I can imagine that for these legacy systems, they are running on very old hardware. I don't know, if one is lucky, Windows 95 or something, there is no easy way to access it. So I find maybe difficult that it's just I come with a USB stick or I can just download it from the internet. What's the process or the rollout for Bevel?

Ali: Yeah, so it's very simple. And so the whole idea was, let's make it very simple for these companies to use it. And secondly, let's make it 100% around data privacy. Because banks, if you have a proprietary code base, you don't want to share it with anyone else, right?

And so the way you do it is very simple. We send you the extension. You download it as one developer. You analyze the code. So it's a VS Code extension. You unzip the file. You run the extension on VS Code, and then you can just share it over Git with all of the other developers in your team. And that's about it.

And the cool thing is nothing leaves your computer. So if you're running a machine, nothing leaves it, not even a single bit. So we also have one version that comes with its own small language model, which means that everything just runs locally on the machine. Or you can just use maybe an Azure hosted LLM if your company has one. Otherwise, everything just stays on the machine, all the code analyzed. We see nothing about it. So it all stays within the company's premises.

Product Development and Feedback

Rod: And now that you're mentioning that so you basically give the extension everything is encapsulated so there is no data transmission you're not really seeing any logs whatsoever of usage here how do you make your product better? So how can you improve it if traditionally something that software has all type of logging and data transfer and so on they are sending so much data to the provider but in your case this is airtight and therefore you do not see anything, but you also do not see how people are using it while they are struggling.

Ali: Yeah, so two things. I think we have to do, or we're doing a lot of customer success, because I mean, already we're startup, right? So we're working very close. So there's two things we're doing. One is just more qualitative slash quantitative feedback. So let's say once a quarter, or I don't know, once every month, we basically work with the developers and we say, OK, what did you struggle with? What was your workflows? All of those things.

And secondly, we store the metadata. So the metadata is nothing to do with the code itself. It's how did they use the application mostly. So it's about the UX of our application, in a way. And we store it on their file, on their computer, right? And we can just request them afterwards. We say, can you look over the file? Look through it. If there is anything proprietary, you can just take it out and just tell us the metadata of how you're using the application, basically.

Programming Language Support

Rod: Are you specialized in any programming language? I can imagine that when one thinks of legacy code, there is so much out there. Have you said, hey, we will just target exclusively couple deployments or maybe like Java legacy deployments or something specific? Or you say, hey, any language we can support?

Ali: Yes, so we are not language agnostic. And the reason for that is every language has its own nuances, its own idiosyncrasies. So that's why we, for every language - so we're using the language server from VS Code. But for every language we have to do some customization. So some things that are just done differently in different languages.

That's the short answer, that every language is different, so we have to have customization. And anything that is agnostic of language is not good for any of those languages because it's very general.

And then what languages we start with? We basically let the market dictate us. We said, OK, where are the people who need us? What languages are they using? And so we got the COBOL, the FORTRAN, and all of those, also the TypeScripts. And we're basically adding support as we go. So as we get customers, it just takes us one or two days to add support for the language and we just are adding more languages and that's our plan. We let the market dictate which languages we are adding support to.

Rod: Which are the most popular ones, which ones are where there is the most demand?

Ali: So it's COBOL especially in the financial industry - mostly like a lot of them run it on COBOL so especially banks and insurances. So I would say that's the bigger one. But also Java is quite around as well.

Company Goals and Funding

Max: Cool. Yeah. I can imagine that there are quite a few of COBOL lying around. They're like those dead corpses in your baskets that you have to take out. I guess just a quick question, Ali, you mentioned that, you know, the goal for the product is what we talk about. And then the goal for the company is to eventually become a place where people want to come and work. I'd love to hear your thoughts, what's your vision on how you can build that and also give us a little bit of sense of where you are from a company perspective, are you guys raising? And just wanted to understand that more too.

Ali: Yeah, of course. I always love talking about the vision. So like I said, the vision is to have this very fulfilling workplace, right? And fulfilling does not mean happy, like always happy and always cheery or some kind of fake positivity. It just means that everyone feels firstly fulfilled. So when they come to work, they're working on something impactful. They're feeling like they basically are using their capabilities to the maximum. So they're utilizing their potential.

And what that means concretely for the company itself is that if you're a manager within the company - we are right now three people, so we are managing ourselves, but once we have more people and you're at a management stage - that your job as a manager is not just are we making money and is the product good and all of those things, but also are the people who are working in your team fulfilled? So we want to make that a key criteria of your management success.

And what that concretely means in terms of values, I think it's because most values are easier said than done and I think a company's values are not what they say, it's what they do when they have to take a decision. So when you have to make a choice, do we do this or this? What do you do, right? So that's why we want to have this mindset. Whenever we come to this fork where we have to make a decision, we will always take the fulfillment of our people that we work with also in consideration and at the top and then everything else after that.

So that's kind of our thinking around it. And we really think that if we have that kind of a culture, that everyone has this - if you know the term, Ikigai - where everyone has this overlap. And you have a team full of those people that you can take on the world. That's our thesis there.

And in terms of where we are right now - so we're actually quite young. We started four months ago full-time and we got what for the Germans they probably know about it - it's called Exist Grundungsstipendium. So it's a scholarship from the government basically so they give you non-dilutive funding for one year to support firstly the founders and then give you some business expenses as well. So we raised 128,000 euros from the government in non-dilutive funding.

So that's how we're running the shop right now. We're in very early stage, so we now have our first pilot customers as well. And we're running the MVP with them. I wouldn't even call it MVP, so we even have a fully functioning product that works. We also have people sign up to our private beta, so we have quite some traction now on the product side as well.

And for fundraising, our goal right now is to after Q1, so after March, have some very clear-cut answers on the product vision and also direction, what kind of product are we building, and does that kind of product and that kind of market actually need a lot of VC funding or not, or do we rather just want to have an angel round and then try to bootstrap it from there. So these questions are still outstanding, and I think it will take us a couple more months to get more clarity on that.

Marketing and Customer Acquisition

Rod: You just mentioned Ali that you may have a waiting list for that. How did that come about? How do people find you? How do you penetrate these customers?

Ali: Yeah, so we started out very enterprise sales like very top-down. But in the end if you do very top-down we have something that the managers really like but no one really uses it on the developer level. And so especially for dev tools it's super important that you have a community around it. Developers they love it. They use it.

And so then we also started doing a lot of product-led growth, right? So we're active on Reddit in the VS Code community, active on LinkedIn. You might have seen some posts from us, also some videos. We're building a community around it. And these people, these developers, they want to use it and they want to use it for their work as well.

And because it comes with a local LLM, so they don't have to actually go to the manager and go through all of those processes. They can just plug and play and use it. And so the goal is let them also use it, give us feedback, and then help us build a case with the managers as well that this is something that's useful and brings value, right? So it's kind of like a sandwich approach of sales, in a way.

Business Model and Pricing

Rod: That's really exciting that you started with an enterprise focus, really targeting decision makers, maybe like CTOs, CIOs, and now you went bottom up, targeting the developers. So how does that impact your business model? Because I can imagine that, for example, in the case of addressing developers, one is charging by the actual usage or maybe by the number of seats. How do you price it?

Ali: Yeah, I think you're spot on with that. So we have to balance the, let's say, community goals with the revenue goals, right? So we have two tiers right now. One of them is, so it's an open core model. For those who don't know what open core is, it means one part of the software, of the solution is free for use for anyone. And that's for individual developers, or a couple of people if they want to use it. And they can just download the extension, use it for free forever.

The second one is the license tier, which is enterprises. And there we have a custom pricing, right now is based on, like firstly, we do this onboarding pilot project, right? Because a lot of it has to customize to their workflow and their code base. And then after that, it's basically a subscription. And right now, we are pricing based on the number of developers. So how many people in the team are using it.

But we will actually switch this pricing sometime in the future, because our thesis is that charging based on the number of developers is a losing battle because the number of developers in one company will for sure go down. So you're basically undercutting yourself in a way. And so we want to transition more to this value-based pricing - how much value we're providing and how much can we actually then skim on top of that. But to really figure out that value, we still have some way to go. So we are slowly transitioning from the seat-based pricing in enterprises to more of this value-based pricing in the very near future.

Closing Thoughts

Rod: If there is a message that you want our audience to get from this conversation, what would that be?

Ali: A lot of messages, but I'll start with the first one is for anyone who works maybe in banks, insurance companies, large enterprises that might suffer from the symptoms that I talked a lot about during our conversation today - if anyone suffers from that, or if you know someone who suffers from that, reach out to me. You can find me on LinkedIn. I think Rod probably will also post a link somewhere or just follow our page. It's called Bevel. And you can just reach out to us. Always happy to talk to you.

And also if you're a developer and you also want to give it a try like I mentioned it's for free so you can just always download the extension. So feel free to reach out to me, always happy to share it with you and walk you through how to use the product.

I think that's the one message and I think a lot of enthusiasts also probably watch the podcast or listen to the podcast. So if anyone is watching and listening to it and is on the fence of whether to start a company or not, I always say this, always start with the why. So if you have a very strong why of you want to start a company then I would highly recommend it. It's a very fun ride. There's lots of ups and downs, but you learn a lot of things. And it's a very transformational journey.

Rod: I must say that here in the show, I know someone who is in the space of needing legacy, support for legacy software, innovation and so on. So you're in the right place, Ali. And if there are others who also feel in the same way, how can they find you? Where can they look for you?

Ali: So like I mentioned, you can follow us on LinkedIn. We are very active there. So we post all the product updates, everything there as well. We will very soon launch on Product Hunt as well. So stay tuned to that. You'll hear about that also on LinkedIn. You can also stay tuned to our website. If you want to join our wait list, there is a link in the website that you can just sign up to. And of course, you can always add me on LinkedIn as well. I'm very happy to stay in touch, talk about anything, also legacy, AI, future software development as well, but also in general. If you want to talk to me about anything career oriented as well, always happy. So feel free to reach out to me on LinkedIn.

Rod: Yes, the URL, the address for anyone who knows is bevel.software, right? So not .com or anything, but .software, the original one.

Ali: Mm-hmm. Yeah. That sounds about right.

Max: I think I really like the vision of you trying to build a company that's fulfilling. One thing that I guess I want to take away is that there are a lot of ways to be successful. We all don't all have to do one way compared to the other, like where the world is going at the moment, as you can probably see in the largest economy in the world. So thank you so much Ali. This has been super insightful on the product as well as the vision of the company.

Ali: This has been super insightful for the product as well as the vision of the company. Thanks a lot, Rod and Max, it was really cool talking to you. So Munich is very early day, and I think also in London. So at least I can say my day is off to very good start. So thank you guys for that.

Rod: Thank you so much for joining us Ali and for everyone else remember to like, subscribe and follow us, join our newsletter, chrisrodmax.com Until next time.

Your Hosts

Follow:

•   Chris: https://linkedin.com/in/christinewang0/

•   Rod: https://linkedin.com/in/rodriveracom/

•   Max: https://linkedin.com/in/maxsontjy/

Social:

• X: https://x.com/chrisrodmax

• Instagram: https://instagram.com/chrisrodmax

• LinkedIn: https://linkedin.com/company/chrisrodmax

• Subscribe: https://chrisrodmax.com

Tags: #chrisrodmax #ai #technews

In this episode, Chris Rod Max discuss the latest trends in AI, focusing on Mistral’s business model, its competition with OpenAI, and the implications of European AI initiatives.

They explore the challenges of AI regulation, the importance of the application layer, and the rise of text-to-web app builders, while also considering the potential impact on developer skills and innovation.

In this conversation, the hosts explore the emerging concept of ‘Vibe Coding’, the implications of Elon Musk’s bid for OpenAI, the challenges posed by deepfakes, and the importance of critical thinking in an era of misinformation.

They discuss how technology is changing the way we learn and interact, the potential consequences of deepfake technology on trust and security, and the need for individuals to verify information in a rapidly evolving digital landscape.

Chapters

• 00:00 Introduction to AI Trends and Mistral

• 01:27 Mistral’s Business Model and Market Positioning

• 06:41 Comparative Analysis of AI Models

• 09:16 European AI Landscape and Nationalistic Push

• 14:20 Regulatory Challenges and Political Dynamics

• 17:05 The Application Layer and AI Development

• 19:47 Text-to-Web App Builders and Their Challenges

• 26:19 The Rise of Vibe Coding

• 29:59 Elon Musk’s Bid for OpenAI

• 39:47 The Deepfake Dilemma

• 49:14 Navigating Misinformation in the AI Era

Takeaways

• Mistral's focus on business and open-source models faces challenges: Despite efforts to differentiate itself from OpenAI and Anthropic, Mistral's impact remains regional, with limited global adoption. Its funding is significantly lower, raising concerns about long-term competitiveness.

• Deepfakes and AI-driven misinformation are rising threats: From financial fraud to manipulated media, deepfake technology is creating security risks. The need for robust identity verification and critical thinking in consuming online content is becoming more urgent.

• AI-powered app builders lower barriers but still require human oversight: While AI-driven development tools accelerate prototyping, they struggle with complex integrations and debugging. "Vibe coding" enables rapid iteration, but over-reliance may erode fundamental programming skills.

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Welcome

Maxson Tee (00:05) Welcome back to another episode of the Chris Rod Max Show. Today we’re joined by Chris and Rod to discuss the latest developments in AI. We’ll be covering several fascinating topics: Mistral’s business strategy, the evolution of web-text-to-app applications, Elon Musk’s attempt to acquire OpenAI, and the growing concerns around deepfakes and AI-generated news distortion. Let’s dive right in.

Mistral’s Business Focus and Strategy

Maxson Tee (01:00) Let’s begin with Mistral’s recent developments. TechCrunch recently published an article about Mistral’s focus on business users. I’ve been following this trend for several months, observing their engagement with banks and organizations to implement Mistral models internally. However, the business model and pricing structure remain unclear. I’d love to get your thoughts on two key aspects: 1. Mistral’s potential business model, especially given Llama’s open-source approach 2. The contrasting strategies between Mistral and competitors like ChatGPT and Anthropic

Rod, as someone who’s been experimenting with these applications, what’s your perspective both as a user and a builder?

Rod (02:34) There are several interesting developments here. Mistral recently launched their mobile application called Le Chat, and social media immediately lit up with praise, particularly from French users. However, the reality is more nuanced - in the UK, for instance, it’s not even in the top 200 most downloaded apps on the App Store, only reaching around top 22 in the productivity category.

Regarding funding, Mistral has raised 1 billion euros, which sounds impressive until you compare it with OpenAI’s funding of approximately 22 billion. Even Anthropic has raised around 14 billion. This funding gap is reflected in model performance - while Mistral’s models are good, they’re now significantly behind competitors like DeepSeek.

Chris (06:41) From an Asian perspective, Mistral isn’t really part of the conversation here. The focus is more on DeepSeek, OpenAI, or Microsoft Copilot. What’s interesting about Mistral’s enterprise approach is its strong French connection - their current business relationships appear to be primarily with French companies.

The recent AI summit featuring President Macron’s support for Mistral seems to indicate a nationalistic push, especially given AI’s increasingly political nature between the US and China. My LinkedIn and X feeds are full of commentary about Europe having another chance in the AI race.

Regulatory Considerations and European AI Strategy

Rod (09:16) I suspect European companies like Mistral will only achieve parity with global players if we see regulations requiring European governmental agencies and large companies to use European models. It could be similar to GDPR, where data must be stored in the European Union - perhaps we’ll see requirements for AI models to be European for use by Europeans.

Maxson Tee (10:01) That’s an interesting perspective. We can see similar approaches in different regions. China, for example, restricted Visa’s entry until UnionPay was established, believing certain infrastructure was too critical to rely on foreign providers. India took a different approach by building more performant systems like UPI and Aadhaar.

The challenge for Europe is coordination across 27 countries. While Mistral may excel in French, handling multiple European languages could be complex.

Text-to-Web App Development

Maxson Tee (19:33) Let’s discuss A16Z’s recent article about using prompts to build products and the rise of AI-powered web app builders. Both of you have entrepreneurial experience - what are your thoughts on how text-to-web app development is evolving?

Rod (19:47) When we look at the most popular app categories online, they’re essentially blogs, shops, CRMs, and ERPs. Despite the variety we see, most applications fundamentally do simple things: present content, manage products, or structure and visualize data. The real innovation here is that when the cost of building software approaches zero, and anyone can build, we might finally achieve Steve Jobs’ vision of computers as ‘bicycles for the mind’ - lean, malleable tools used efficiently for specific purposes.

OpenAI Acquisition Attempt

Maxson Tee (30:54) Let’s discuss Elon Musk’s attempt to buy OpenAI. He’s leading a consortium offering $97 billion, while OpenAI is seeking $40 billion at a $300 billion valuation. Sam Altman’s response on X was simply “no thanks,” with a counter-offer to buy X for $9.7 billion. What are your thoughts on Musk’s stated intention to keep OpenAI non-profit?

Rod (32:07) This situation can be viewed from multiple angles. Looking at OpenAI’s current structure, a software engineer’s total compensation can reach 1.2 million per year, including stock grants. This requires significant cash flow and constantly increasing valuation to maintain such compensation packages. It’s challenging to sustain this in a non-profit model.

Deepfakes and AI Security

Maxson Tee (40:45) Let’s address the growing concern about deepfakes, sparked by Scarlett Johansson’s recent call for government regulation. Gartner predicts that by 2026, 30% of enterprises will consider identity verification and authentication unreliable due to deepfakes. How are your networks responding to this threat?

Rod (41:09) This isn’t entirely new - we’ve dealt with photo manipulation since the early internet days. However, the current situation is more dangerous because we can now bypass security checks and identification systems. We might need to return to offline processes for certain verifications, similar to how bank accounts once required in-person verification.

Chris (42:54) I expect we’ll see much more emphasis on two-factor authentication and cybersecurity measures. Business operations will likely become more complex - think about how we now need to sign into Microsoft every few days. Universities are already working on PhD topics about securing against deepfakes and developing AI-generated content detection.

Closing Remarks

Maxson Tee (51:50) Thank you everyone for tuning in. Remember the 80-20 rule we discussed: AI can handle 80% of the work, but you need to know your goals for the remaining 20%. We’ve covered Mistral’s focus, text-to-app development, Musk’s OpenAI bid, and the challenges of deepfakes and AI-generated news. Join us next time for more insights into the evolving world of AI.

Rod (52:48) You can find us at chrisrodmax.com

Your Hosts

• Follow: 

  • Chris: https://linkedin.com/in/christinewang0/ 

  • Rod: https://linkedin.com/in/rodriveracom/ 

  • Max: https://linkedin.com/in/maxsontjy/ 

• Social: X: https://x.com/chrisrodmax 

• Instagram: https://instagram.com/chrisrodmax 

• LinkedIn: https://linkedin.com/company/chrisrodmax 

• Subscribe: https://chrisrodmax.com 

In this episode of the Chris Rod Max Show, Sam Miller from Artanis shares his journey from an economist at the Bank of England to an AI entrepreneur. He discusses his passion for economics and how it led him to explore AI, eventually founding a fitness AI company.

Sam delves into the challenges of building AI, the transition from an agency model to a product-focused approach, and the importance of data labeling in creating effective AI solutions. He emphasizes the need for good work, reputation building, and the balance between service and product in the AI landscape.

In this conversation, Sam discusses the evolution of AI labeling tasks, emphasizing the shift from unambiguous labeling to subjective interpretations. He introduces the concept of ‘policy’ in AI, which aims to provide clear instructions for labelers to ensure consistent data labeling.

Sam also shares insights about his company, Artanis, which focuses on building trustworthy AI products. The discussion touches on the current state of the AI market, the importance of policy in medical AI applications, and the dynamics of investor relationships.

Sam concludes with thoughts on the future of AI and the significance of honesty in business practices.

Chapters

• 00:00 Introduction to AI and Personal Journey

• 03:12 Transition from Economics to AI

• 06:13 Starting a Fitness AI Company

• 08:57 Building AI: The Process and Challenges

• 12:08 From Agency to Product Company

• 15:04 Navigating the AI Agency Landscape

• 18:11 Scaling AI Services vs. Products

• 21:07 The Importance of Data Labeling in AI

• 23:52 Core Principles of Building Effective AI

• 25:57 The Evolution of AI Labeling Tasks

• 29:26 Defining Policy in AI Labeling

• 31:15 Artanis: Building Trustworthy AI Products

• 33:27 Current State of AI Market

• 36:34 Policy Implementation in Medical AI

• 40:14 Understanding Investor Dynamics

• 43:42 The Importance of Truth in Business

• 44:57 Future Prospects of AI in Everyday Life

Takeaways

1. Exercise Caution with Current AI Solutions Sam explicitly warned: "AI solutions don't work perfectly right now. Don't buy one if someone's trying to sell you one." This is a crucial red flag for enterprises to be skeptical of vendors promising fully automated AI solutions. The emphasis on human-in-the-loop implementations suggests that enterprises should expect to maintain significant human oversight of any AI systems they deploy.

2. Focus on Specific Policies Rather Than Vague Capabilities When evaluating AI solutions, enterprises should look for vendors who can articulate specific, concrete policies for their AI systems rather than making broad claims. As Sam illustrated, vague instructions like "be the world's best doctor" are less valuable than specific policies like "tell users who are having a stroke to go to the hospital." Enterprise buyers should demand clear, testable specifications for AI behavior.

3. Look for Natural Language Interface Potential Sam highlighted that in the next five years, AI will transform the human-computer interface to natural language, "opening up opportunities for people who don't like traditional computer interfaces to get more value from software." For enterprises, this suggests evaluating AI solutions not just for their core functionality, but for their potential to make complex systems more accessible to a broader range of employees through natural language interactions.

YouTube Episode

Spotify Podcast

Episode Transcript

Maxson Tee: Welcome to the Chris Rod Max Show. Today we have a special guest, Sam, joining us along with our co-host Rod. Chris is also with us from a café somewhere. Sam, great to have you here.

Sam: Thanks! Yes, I'm currently at ***, though I'd almost rather be in that café given what these folks know about AI [laughs]. But I'm just kidding - they're really smart people here. I want to give a shout-out to Boulderton because they've generously given us free space for six months. We've met great people and gotten excellent advice, even though we occasionally have different views on AI.

Career Journey: From Economics to AI

Maxson Tee: Perfect! Let's start with your background. You went from being an economist at the Bank of England to working in AI. Could you walk us through that journey?

Sam: I studied economics at university and really loved it. I wasn't particularly motivated by money - I wanted to have a macro-scale impact on society and policy. The Bank of England seemed like a good way to do that. I thought, 'If I could tweak the right knob for the government, I could impact the whole society.'

The reality of policymaking as a 22-year-old was different, of course. You're not the one tweaking the knobs - you're writing documents that people usually ignore. But half my job was research, which I really enjoyed. I had the unusual opportunity to write economic research papers without a PhD, thanks to a supportive manager.

Working in the public sector meant the job wasn't too intense, so I started learning Python as a hobby. The first time I wrote a for loop, I was amazed - 'Wow, it can do a million things in a second!' I got into machine learning initially just to improve my Python skills.

My background in econometrics, which is essentially applied statistics, had significant overlap with machine learning. This led me to apply for a machine learning AI PhD at the Alan Turing Institute. They make exceptions for people without CS backgrounds, and thanks to my published research, I got in. The transition from economics to AI isn't as dramatic as it might seem - economics and econometrics are actually great backgrounds for AI, though you do need to learn programming.

First Startup Experience

Maxson Tee: After your PhD, you started a company, right?

Sam: Actually, let me correct that - I started my first company nine months into my PhD, not after it. It's been over seven years now, so I can talk about it freely.

Maxson Tee: Tell us about that venture. Was it related to fitness?

Sam: Yes, but in an interesting way. I used to compete in powerlifting during university - bench press, squat, deadlift, lifting as heavy as possible. I once burst a blood vessel in my eye during a squat competition, which the other lifters thought was pretty impressive!

I enjoyed teaching people about weightlifting and coached many friends and colleagues from the Bank of England through WhatsApp. They'd send me videos of their form, and I could always spot how to improve it.

At the Turing Institute, I sat next to someone running a computer vision lab working on pose estimation, or skeleton tracking - determining body coordinates from images or videos. I realized my weightlifting coaching was essentially about body coordinates: 'hips need to be lower than knees at the bottom of the squat.' We could create an AI version of my WhatsApp coaching - an AI personal trainer to solve the problem of poor form in the gym.

Building the First AI Product

Maxson Tee: That's fascinating! Can you tell us about the process of building that AI?

Sam: The technology worked quite quickly because there had been significant advances in pose estimation. We could use pre-trained models rather than training our own. When we got it working fast, I thought, 'We're going to be billionaires! We have fancy technology and a great idea!'

My co-founder had a strong background as a strategy consultant from one of the big consultancies. When he quit his job to join, people from his previous startup invested, and we quickly raised an angel round based mainly on his connections. We thought it would only take six months to build. That was Atlas.

The Agency Journey

Rod: You've been working as an agency, and now you're transforming into a product company. What's the current situation with AI agencies? There's discussions about agencies achieving product-like margins through automation.

Sam: The way we kept Atlas going for four or five years was interesting. We raised about 500K over four years - not much in the tech world for a team of five. Because my colleague and I both had AI PhDs, companies would occasionally email us after seeing Atlas, not wanting to buy our product but wanting to pay us to build their AI.

Initially, we would decline because we were focused on the fitness app. But when money got tight, we started taking these projects to fund Atlas. Eventually, we realized we'd pivoted far from our original vision - we'd gone from being powerlifters making a strength training app to counting star jumps in living rooms.

Building Trust in AI

Sam: The key problem in AI right now is trust. Most AI being deployed is human-in-the-loop because people don't actually trust that it works. At Artanis, we've developed something we call 'policy,' which we believe is the missing core of the AI stack.

Medical triage as an example. If you have a bot deciding whether people should go to the hospital, different doctors might have different policies. You can't start with vague instructions like 'You're the world's best doctor, answer accurately and reliably.' You need specific policies: 'Tell users who are having a stroke to go to the hospital.'

The Future of AI

Maxson Tee: What excites you most about AI in the next five years?

Sam: First, let me caveat: AI solutions don't work perfectly right now. Don't buy one if someone's trying to sell you one. But in five years, I think AI agents will work - you'll be able to give high-level instructions like 'book a meeting with Rod next week' and the agent will handle all the details.

This is compelling because it changes the human-computer interface to natural language, opening up opportunities for people who don't like traditional computer interfaces to get more value from software.

Closing Thoughts

Rod: Where can people find you and learn more?

Sam: You can visit artanis.ai to subscribe to our content on Substack or our monthly updates. You can also email me directly at sam@artanis.ai - if I get too many emails to respond to, that's a great problem to have! We'd love to hear any feedback, including thoughts on this podcast.

Rod: Thank you, Sam, for joining us. Everyone, don't forget to like, subscribe, and check out ChrisRodMax.com for our newsletter.

Sam: Thank you guys, appreciate it.

Maxson Tee: Thank you.

Your Hosts

• Christine Wang: LinkedIn

• Rod Rivera: LinkedIn

• Maxson J.Y. Tee: LinkedIn

Follow

• Chris: LinkedIn

• Rod: LinkedIn

• Max: LinkedIn

Social

• X: ChrisRodMax

• Instagram: ChrisRodMax

• LinkedIn: Company Page

• Subscribe: Website

Tags: #chrisrodmax #ai #technews

In this episode, Chris and Rod discuss the evolving landscape of AI, focusing on recent model releases, the geopolitical implications of AI development between the US and China, and the push for AI-first strategies in organizations.

They explore the confusion surrounding AI model names, the role of companies like Palantir in government AI adoption, and the differing approaches to AI in regions like Singapore and India. The conversation highlights the competitive nature of AI innovation and the underlying mistrust that shapes the current digital landscape.

Chapters

• 00:00 Introduction to AI Controversies

• 00:55 Recent AI Model Releases

• 03:30 The Race for AI Specialization

• 06:08 Geopolitical Tensions in AI

• 11:06 Digital Warfare and Mistrust

• 15:25 The Push for AI-First Organizations

• 18:01 Singapore’s AI Strategy

• 19:20 India’s Emerging AI Landscape

Takeaways

AI Model Confusion and the Shift to Specialization

The AI landscape is getting harder to navigate, with new models constantly emerging under cryptic names. Most users struggle to differentiate them, and without clear guidance, picking the right one feels like trial and error. Meanwhile, China is leading a shift toward specialized AI models, focusing on niche enterprise applications rather than broad, general-purpose AI.

Geopolitics Is Reshaping AI Adoption

AI isn’t just about innovation—it’s a battleground for influence. Palantir’s rejection of Chinese AI models reflects growing national security concerns, and countries like Australia and Taiwan are already banning certain models. With deep mistrust on both sides, AI development is increasingly shaped by political lines rather than just technological progress.

AI-First Strategies Are Reshaping Government and Business

The push for AI-first organizations is gaining momentum. The US government is looking to streamline operations with AI, treating it more like a startup would. Singapore is actively investing in AI talent and enterprise adoption, while India, though not yet a major player in foundational AI, is leveraging its strength in IT services to drive digital transformation. The AI race isn’t just about better models—it’s about who can integrate them most effectively.

YouTube Episode

Spotify Podcast

Episode Transcript

Rod: Welcome to another episode of the Chris Rod Max show. We are the AI’s buyer briefing. Hello Chris, how are you?

Chris: Fantastic! I’m excited to be here on the show again today to discuss the fascinating controversy surrounding the US-China AI competition.

Episode Overview & Announcements

Rod: Excellent! We have several engaging topics to cover and, as usual, limited time. But before we dive in, I’d like to remind our audience to like, subscribe, and join our newsletter at ChrisRodMax.com. There you’ll find all our latest updates, recordings, and episodes.

Recent AI Model Releases

Rod: Let’s start with last week’s exciting releases from OpenAI and Mistral. Mistral has introduced Mistral Small 3, a highly performant yet compact model tailored for enterprise use cases like detection, healthcare, and robotics. OpenAI has released two offerings: one is more of a feature than a model, and the other is the O3 Mini model. O3 Mini focuses more on reasoning capabilities rather than traditional question-answering, distinguishing it from models like DeepSeek R1. They’ve also introduced a deep research feature, similar to Google’s Gemini, which helps prepare research briefings for specific topics. Chris, have you had the chance to experiment with these models?

Chris: Honestly, it reminds me of iPhone or iPad releases – there are many similar models with subtle differences. For the average user, it’s quite challenging to distinguish between them. The performance differences mainly matter to engineers and tech enthusiasts who actively compare and optimize their usage. Most general users don’t delve deep enough into prompting or questioning to notice significant differences.

What I find particularly interesting is the market’s reaction to DeepSeek and this trend toward specialization. We’re moving away from the initial approach of creating massive, general-purpose models toward building specialized, niche models trained on specific sector data. This shift represents a significant trend in the industry.

User Experience with Different Models

Rod: When you use platforms like ChatGPT or Claude, they offer options to switch between different models. Do you actually use these options? Do you stick with the standard model?

Chris: I rarely switch between models because the naming conventions have become quite confusing – O3, O3 Mini, OpenAI Mini, and so on. For general use cases like polishing emails or summarizing information, any of these models performs adequately. The differences don’t significantly impact my workflow.

Rod: I occasionally switch models, but I agree that the naming is cryptic and confusing. There’s no clear guide that says, ‘for this specific problem, use this specific model.’ It’s more about experimentation and potentially returning to a different model if the first choice doesn’t work well. The market seems divided between Western AI and Chinese AI, with everything falling into one of these two categories.

Integration and User Interface

Chris: Exactly, and to add to your point, we discussed last week that the real differentiator isn’t necessarily the foundational model – it’s who owns the interface to the end user and how well it integrates into existing workflows. Humans prefer convenience, so seamless integration often matters more than a 1% performance improvement.

The China-US AI Divide

Rod: This brings us to the broader discussion about Chinese models and their adoption. Let’s talk about Palantir, a data company that provides bespoke data pipelines for government and enterprise clients, and has increasingly transformed into an AI solutions provider. Their CEO, Alex Karp, known for controversial statements, recently made headlines regarding DeepSeek R1. He stated that he doesn’t foresee adoption of this Chinese model in the Western world and discourages US customers from using it. Chris, what’s your take on these statements? Should governments avoid Chinese AI entirely?

Chris: Don’t you think this has essentially become a digital war? It started in 2010 when China blocked Western companies like Facebook and Google, and now it’s evolved into an AI arms race. While the US limits hardware supply to China, Chinese entrepreneurs have shown remarkable innovation in working around these restrictions.

I think we need to distinguish between two aspects here. First, this competitive tension is driving innovation, pushing both sides to develop more specialized and efficient models. That’s healthy for the industry. Second, Palantir’s recommendation against using DeepSeek in Western governments reflects broader geopolitical concerns. It’s similar to suggesting we shouldn’t use any Chinese products in the West, or vice versa. While there are always surveillance risks – and honestly, even the US probably monitors our conversations – it’s a complex issue.

International Restrictions and Trust

Rod: You’re right about the 2010 turning point. China’s increasing restrictions on internet services and international providers have been used by Western sources to justify similar restrictions. We’re now seeing countries like Australia, Taiwan, and Italy forbidding the use of DeepSeek R1 in government agencies, primarily citing user data and privacy concerns, but likely also to avoid conflicts with the US administration.

Chris: This really does feel like a modern digital war driven by deep mistrust. While I can’t verify the surveillance risks, everyone’s extremely cautious about potential data leaks or security breaches. There were some interesting memes about DeepSeek’s responses regarding Taiwan’s status, which people saw as evidence of bias. It demonstrates how different cultural perspectives shape these AI systems. While everyone should be free to choose their preferred model, there’s a larger problem of international mistrust at play.

Different Standards for AI Models

Rod: It’s fascinating how we hold different AI models to different standards. People criticize DeepSeek’s responses to questions about Chinese history or politics, but we don’t apply the same scrutiny to Western models’ limitations. For instance, OpenAI’s image generator won’t create images of Donald Trump or trademark-protected content like Mickey Mouse characters.

Chris: Exactly – it’s a form of censorship, albeit with different intentions like data privacy protection. We just don’t typically view Western restrictions through the same lens of censorship.

Business Impact and Government Adoption

Rod: The business implications are significant. Palantir’s stock rose 24% after their recent quarterly report, showing strong market support for their position as a government-approved AI provider.

Speaking of government adoption, Elon Musk’s role in the US administration with his efficiency drive is promoting an AI-first strategy for governmental agencies. Chris, do you think organizations will embrace this AI-first approach?

AI-First Strategy and Implementation

Chris: Implementing an AI-first strategy in existing organizations is challenging due to legacy systems and traditional mindsets. However, having business leaders in politics might help drive efficiency improvements. While the concept of an AI-first strategy remains somewhat vague, we’re at a watershed moment where governments are actively engaging with AI’s potential and developing ethical frameworks for their countries.

Singapore’s Perspective

Rod: Given your base in Singapore, known for running government like a business, how is this AI-first wave being received there?

Chris: Singapore is quite proactive in this space. There’s a major push toward AI education and talent development. While there’s some research into foundational models, the focus is primarily on practical applications. Many organizations are exploring AI implementation, often relying on established B2B players like Microsoft Copilot. The proximity to China has also generated significant interest in DeepSeek.

Rod: Interestingly, 22% of Nvidia’s revenue comes from Singapore, leading to discussions about whether Singapore serves as a gateway for China to access restricted technologies and hardware.

India’s Role in AI Development

Chris: What about India? Given their excellent engineers, it’s surprising we haven’t heard about any foundational models from India.

Rod: While there might not be much development at the foundational level, there’s a significant trend of Indian engineers gaining experience in the West before returning home to leverage growth opportunities. India’s strength lies more in IT services and digital applications than in foundational AI research. The country has numerous large companies and a robust ecosystem across multiple cities. It might be just a matter of time before India follows China’s path, requiring companies like OpenAI to establish local offices and conduct research within India.

Conclusion

Rod: To summarize this week’s discussion: 1. The AI model landscape is becoming increasingly complex with cryptic naming conventions and unclear use cases 2. Palantir is positioning itself as a government-safe AI provider while discouraging Chinese AI adoption 3. Government organizations are considering startup-like AI-first approaches following US administrative changes

Remember to tune in next week for another episode of the Chris Rundach Show and join us at chrisRodmax.com. Until next time!

Your Hosts

Follow: 

• Chris Wang: https://linkedin.com/in/christinewang0/ 

• Rod Rivera: https://linkedin.com/in/rodriveracom/ 

• Max Tee: https://linkedin.com/in/maxsontjy/ 

Social: 

• X: https://x.com/chrisrodmax 

• Instagram: https://instagram.com/chrisrodmax 

• LinkedIn: https://linkedin.com/company/chrisrodmax 

• Subscribe: https://chrisrodmax.com 

Tags: #chrisrodmax #ai #technews

Vitalii shares insights from his extensive experience in data science and machine learning, emphasizing the need for persistence and adaptability in the rapidly evolving AI landscape.

Chapters

• 00:00 Introduction to Dynamiq and AI Development

• 04:07 Vitalii’s Background and Experience in AI

• 09:29 Challenges in Deploying Generative AI Solutions

• 13:54 Success Stories: Real-World Applications of Dynamiq

• 16:51 Pitching Dynamiq’s: Differentiation in a Crowded Market

• 22:23 Understanding Customer Needs and Education

• 26:17 Addressing AI Security and Compliance Concerns

• 30:16 Dynamiq’s Unique Features and Capabilities

• 36:15 Platform Walkthrough: Demonstrating Dynamiq’s Functionality

• 45:02 Key Learnings and Advice for Founders

Takeaways

• Dynamiq’s simplifies the journey from prototyping to deployment.

• The market for generative AI development tools is growing.

• New tools are needed for generative AI applications.

• Dynamiq’s aims to unify various functionalities in one platform.

• Companies can deploy features quickly using Dynamiq’s.

• Customer education is essential for understanding AI capabilities.

• Data privacy and security are critical concerns for enterprises.

• Dynamiq’s offers an end-to-end solution for AI development.

• Technical capability is necessary to use Dynamiq’s effectively.

• Persistence and adaptability are key for success in AI.

YouTube Episode

Spotify Podcast

Episode Transcript

Chris: Welcome to another episode of the Chris Rod Max show, where we discuss the latest AI developments with industry leaders. Today, we’re joined by my co-hosts Rod and Max, along with our special guest, Vitalii Duk, CEO and founder of Dynamiq’s.

Vitalii: Hey guys, thanks for having me.

Understanding Dynamiq’s: The Enterprise GenAI Platform

Chris: Let me share my understanding of Dynamiq’s - it seems like a ‘what you see is what you get’ toolkit for AI deployment. Similar to how non-technical users can build websites by moving boxes around, Dynamiq’s appears to let users deploy AI products by linking databases in a low-code manner. But Vitalii, please tell us more about yourself and Dynamiq’s.

Vitalii: I’m Vitalii, founder and CEO of Dynamiq’s. My background spans over 11 years in data science. Dynamiq’s is an end-to-end platform for building GenAI applications. Our core focus is simplifying the entire journey - from prototyping to deployment, observability, and implementing guardrails. We have a low-code builder at the platform’s heart, with pre-built blocks for common tasks like vector database connections. We also support custom Python code integration for engineers and data scientists who prefer more control.

Background and Industry Experience

Chris: Before we dive deeper into Dynamiq’s, could you share more about your background and how you arrived at this idea?

Vitalii: I’ve been deeply involved in building internal machine learning platforms. I worked at Careem, later acquired by Uber, where I led the machine learning platform development across 14 countries serving 50 million users. We built everything from A-B testing platforms to feature stores. This experience, plus collaboration with Uber’s ML team, gave me extensive insights into building platforms.

Challenges in Enterprise GenAI Adoption

Chris: What challenges do enterprises face when adopting generative AI solutions?

Vitalii: There’s been a paradigm shift in AI applications. Previously, we built specific models for individual tasks like sentiment analysis or entity extraction. Now, foundation models can handle multiple tasks, with even more complexity through agentic capabilities. This new paradigm requires new tools focused on prompts, LLM fine-tuning, and RAG techniques. The traditional ML tooling doesn’t fit this new world of LLM-centric applications.

Platform Development and Feature Prioritization

Rod: How did you decide which features to build first, given the platform’s extensive functionality?

Vitalii: We had a clear vision from early customer discovery calls about core platform needs. While not everything existed in our original vision - like LLM agents and RAG capabilities which came later - we knew we wanted to unify everything under one roof. The market is currently fragmented with separate tools for observability, evaluations, and RAG. Our vision was to create a unified platform to eliminate the complexity of stitching together multiple tools.

Customer Success Story

Chris: Could you share a success story of how a company has used Dynamiq’s?

Vitalii: We have a Series B B2B SaaS platform customer who’s built 5-6 different use cases on Dynamiq’s. They’ve implemented everything from document summarization to customer-facing assistants and internal HR co-pilots. They deployed these features in hours instead of months, which was crucial given their limited engineering resources. They get built-in observability, evaluation capabilities, and the flexibility to test different models like OpenAI or Anthropic.

Market Education and Differentiation

Rod: How do you present Dynamiq’s to companies given the market confusion around AI tools?

Vitalii: One surprising insight is that many companies aren’t as advanced in GenAI as we might think. Some are just looking for centralized enterprise access to ChatGPT. We often need to educate customers about what’s possible with LLMs and how they can streamline operations. Our pitch focuses on showing potential use cases and demonstrating how quickly they can be implemented compared to building from scratch.

Security and Compliance

Chris: How do you address AI security and compliance concerns?

Vitalii: Data privacy and sovereign AI are hot topics. We’ve built much in-house and maintain infrastructure agnosticism. We offer on-premise deployment options in private clouds or bare metal servers, ensuring data stays within customer premises. Customers can deploy private instances of open-source models like Llama 2 and use vector databases like weaviate, all within their infrastructure. However, we remind customers that running LLMs on private infrastructure can be costly.

Platform Demonstration

Rod: Could you show us how Dynamiq’s looks?

Vitalii: [Demonstrates platform features] - Low-code builder with pre-built templates - RAG workflow capabilities with query refinement - Multiple LLM integrations (OpenAI, Anthropic, open-source models) - Vector database integrations - Agent definition and tool configuration - Multi-agent orchestration - Python customization options - Monitoring and observability features - Model comparison and evaluation tools

Closing Advice

Chris: What have you learned in your year in this space?

Vitalii: Persistence is crucial for founders in this space. Despite the crowded market, long-term focus and customer engagement are key. You need to iterate quickly as the technology evolves - for example, we initially focused heavily on fine-tuning, but now see more demand for RAG and agentic solutions. Being adaptable to customer needs is essential in this rapidly changing field.

Contact Information

Chris: How can listeners reach you?

Vitalii: Find me on LinkedIn as Vitalii Duk or visit https://www.getdynamiq.ai/ to book a demo.

Chris: Thank you for joining us. Listeners, please subscribe to our newsletter and follow our channels.

Vitalii: Thank you, see you, bye bye.

Your Hosts

Christine Wang Rod Rivera Maxson J.Y. Tee

In this episode, the hosts discuss the recent launch of DeepSeek, a groundbreaking AI model with significant implications for the tech industry and geopolitical landscape. They explore its technical innovations, the market’s reaction, and the potential for smaller players to compete in AI development.

The conversation also delves into the importance of trust in AI models, the future of AI innovation, and insights for entrepreneurs navigating this rapidly evolving space.

Chapters

• 00:00 Introduction and Overview of DeepSeek

• 02:01 DeepSeek’s Technical Innovations and Implications

• 05:48 Geopolitical Context and Market Reactions

• 09:51 Adoption and Trust in AI Models

• 13:55 The Future of AI Development and Innovation

• 18:06 Entrepreneurial Insights and Market Dynamics

Takeaways

1. DeepSeek’s Breakthrough Challenges AI Development Norms

   DeepSeek R1 marks a significant shift in AI by eliminating human involvement in training through a teacher-student model. This challenges traditional reinforcement learning with human feedback (RLHF) and showcases that high-quality AI can be developed with minimal manual oversight.

2. Geopolitical and Market Disruptions

   The launch of DeepSeek has had major geopolitical and economic implications, causing a drop in NVIDIA’s stock and raising concerns in the West about China’s growing AI capabilities. The open-source nature of DeepSeek also introduces new dynamics in AI accessibility and competition.

3. Trust and Cost Efficiency Shape AI Adoption

   While DeepSeek offers powerful capabilities at a fraction of the cost of Western models, enterprise adoption depends on trust, security, and regulatory considerations. Open-source AI democratizes access, but businesses must balance innovation with compliance and practical implementation.

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction & Welcome

Maxson Tee: Welcome back to another episode of the CRM show, the Chris Rod Max show. We have my co-host with me today, Chris Rod. Welcome back.

Chris: Hey guys! Hi!

Rod: Hi everyone.

Setting the Stage: Recent Developments

Maxson Tee: This week has been packed with significant developments. Following last week’s inauguration, we witnessed the launch of DeepSeek, which we’ll discuss in detail today. Interestingly, this led to some controversy in the West, with OpenAI claiming DeepSeek trained its model using OpenAI’s model. We’ve also seen interesting discussions about ChatGPT rappers, which were previously criticized but are now gaining acceptance. Let’s dive into these topics.

The biggest story this week is definitely DeepSeek. It’s notable that it caused a 600 million or billion drop in NVIDIA’s value. Rod, I see you’re ready to comment - what makes DeepSeek so special?

DeepSeek’s Revolutionary Impact

Rod: Max, while you’re calling this the news of the week, I’d go further and say this could be the news of the year. The impact is comparable to a meteorite hitting Earth - that’s how significant this is.

Maxson Tee: That’s a big claim, considering we’re just at the start of the year.

Rod: Indeed, it’s been an eventful year, but I believe when we look back at the end of 2025, DeepSeek R1 will stand among the most relevant and impactful developments. While there’s a geopolitical aspect we could discuss, I’ll focus on the technical significance: DeepSeek has demonstrated, for the first time, that you don’t need human involvement to create better models. Historically, with OpenAI models, human feedback was essential - indicating good or bad answers, guiding the model’s direction. DeepSeek R1 achieved its results through a teacher-student model approach, eliminating the need for human review and assessment.

Non-Technical Implications

Maxson Tee: Interesting. Chris, from a non-technical perspective, what are your thoughts on DeepSeek’s implications?

Chris: As someone who does a lot of hiking and trail running, I’ve been listening to numerous podcasts, and this is undoubtedly the top news right now. What’s fascinating is how people are trying to dissect DeepSeek, searching for potential flaws or questioning its legitimacy. The remarkable fact is that it’s legitimate and achieved at a cost of $6 billion versus roughly $100 billion for comparable models.

I think the market’s reaction reveals significant fear, especially given the current geopolitical dynamics between the West and China. While many are surprised that China has developed a model matching Western capabilities, this development isn’t entirely unexpected. The real question now lies in adoption and future user implementation.

Market Impact and Model Commoditization

Maxson Tee: Absolutely agree, Chris. We’re witnessing a commoditization of AI models, with Alibaba and others releasing their own versions claiming superiority. The key factor becoming increasingly important is trust in these models and their adoption over time.

Referencing Nassim Taleb’s recent Bluebook interview, he pointed out that while many innovators create initial technologies, the early innovators often don’t capture most of the value. Instead, value comes from adoption and building valuable applications on top of the technology.

Rod, from your technical perspective, given your experience in building AI models, how do you view the balance between trust issues and cost implications?

Technical Implementation and Accessibility

Rod: We need to distinguish between two aspects. First, there’s the web-based chatbot similar to ChatGPT, which raises data hosting concerns since it’s based in China/Hong Kong rather than the US or Europe. This might limit its use for certain industries or companies.

However, it’s also an open-weights model, meaning anyone can download and run it locally without sharing data externally. Remarkably, estimates suggest you can build a $6,000 gaming-level computer capable of running their most powerful model at full capacity. While it might be slightly sluggish at eight tokens per second, it’s still productive - a dramatic shift from requiring billion-dollar infrastructure.

Maxson Tee: That’s fascinating - it’s similar to downloading Bitcoin’s code to run your own mining system. Now you can do the same with AI models at home.

China’s Approach to AI Development

Chris: This connects to our earlier discussions about China’s approach to AI. It’s remarkable that DeepSeek’s model is open source, allowing anyone to download and contribute to its development. We previously discussed how China builds AI solutions vertically by industry, fostering collaboration on datasets and training within specific sectors like shipping or banking. Now we’re seeing these approaches converge.

Innovation and Global Competition

Maxson Tee: There’s an interesting irony here - China, typically seen as more centralized, is taking an open-source approach, while the West, traditionally more distributed, has concentrated AI development among a few players. DeepSeek, coming from a hedge fund rather than a major tech company, demonstrates interesting potential.

This shows that smaller players can compete effectively - it’s not just about money and computing power. Countries like the UK and France, with sufficient intellectual capital, could potentially develop similar models. This creates more opportunities for innovation across different countries and better pricing from an economic perspective.

Enterprise Adoption and Implementation

Rod: Max, given your perspective on both enterprise and startup sides, how are companies approaching implementation? Are they actively exploring ways to benefit from it, or are they hesitant due to its foreign origin?

Maxson Tee: From a model perspective, I’m confident many AI departments are examining DeepSeek due to its open-source nature. In financial services, many technologies are derived from open-source software, including risk software. When an AI model is fully open source and trustworthy, adoption typically follows.

Regarding use cases, banks are showing flexibility about which models they use - they care more about results and cost efficiency than the specific model. This cost aspect is crucial for growth, as lower costs enable more innovation attempts and higher chances of breakthrough successes.

Market Response and Technical Innovation

Chris: The impact of DeepSeek varies by audience. We saw significant stock market drops in technology and semiconductor sectors, partly because DeepSeek doesn’t require the latest GPU chips. However, for enterprises, the specific model matters less than its effectiveness.

From a research perspective, it’s interesting how export restrictions may have pushed China to develop more creative, cost-effective solutions. notably, DeepSeek seems to have overcome data limitations through innovative self-learning algorithms.

Technical Details and Market Implications

Rod: Regarding the teacher-student model approach, there are rumors that DeepSeek might have used OpenAI’s models as teachers, though this isn’t confirmed and would violate OpenAI’s terms of service. Some users report the model occasionally identifying itself as an OpenAI model, suggesting possible training connections.

Regarding market impact, Nvidia’s stock remains down about 15% since this news. This raises questions about future GPU requirements and the entire AI development model, including initiatives like Stargate’s $500 billion investment.

Future Implications and Applications

Maxson Tee: The requirements really depend on the goal. Building AGI might need massive computation, but solving daily problems often doesn’t require top-end capabilities. The market reaction follows typical patterns - initial excitement, potential exuberance, then adjustment.

Regarding applications, Greg Eisenberg’s observation about ChatGPT wrappers is interesting - what was initially criticized as simple repackaging now appears to be an effective approach, providing direct solutions to specific customer needs with commoditized models.

Closing Thoughts on Implementation

Chris: It’s ultimately about the interface - creating something people want to integrate into their daily lives. The goal is to own real estate on people’s phones, becoming their go-to solution, regardless of the underlying engine.

Maxson Tee: Exactly - like transportation, users care about getting from A to B, not the specific vehicle.

Rod: From an enterprise perspective, the real challenge is implementation - ensuring frontline workers can use this technology effectively. While AI might threaten certain consulting roles, the moat lies in relationships and system integration. The advantage comes from providing this connection and maintaining customer relationships.

Maxson Tee: Indeed - “whoever controls your distribution controls your life.” This presents opportunities to rethink software deployment, with AI potentially helping bridge legacy systems. We’ll continue this discussion next week. Remember to like and subscribe, and share your thoughts about DeepSeek in the comments.

Your Hosts

Christine Wang Rod Rivera Maxson J.Y. Tee

Follow Chris Rod Max:

https://linkedin.com/in/christinewang0/

https://linkedin.com/in/rodriveracom/

https://linkedin.com/in/maxsontjy/

Follow on X:

https://x.com/chrisrodmax

Follow on Instagram:

https://instagram.com/chrisrodmax

Follow on LinkedIn:

https://linkedin.com/company/chrisrodmax

Subscribe to the newsletter:

https://chrisrodmax.com

#chrisrodmax #ai #technews

In this episode, we discuss the recent announcement of a $500 billion AI fund backed by the Trump administration, focusing on its implications for the AI landscape, particularly in relation to OpenAI’s closed model and the competition with China’s open-source strategies.

Subscribe now

We explore the political dynamics influencing these developments and the potential impact on the market and innovation.

Chapters

• 00:00 Introduction to AI Developments

• 01:47 The $500 Billion AI Fund

• 06:09 OpenAI’s Closed Model and Global Competition

• 10:27 China’s Open Source Strategy

• 16:09 The Political Landscape and Market Reactions

Takeaways

1. The $500 Billion AI Fund and US Strategy: The Trump administration's $500 billion AI fund aims to solidify the US as a global leader in AI by investing heavily in data centers and infrastructure. This initiative signals that AI-focused companies may have higher chances of success if they align with US priorities, emphasizing the administration's goal to centralize AI innovation in America.

2. Global Competition Intensifies**: While the US moves toward centralizing AI development, China is adopting an open-source approach, with companies like DeepSeek offering free, competitive AI models. This shift highlights a growing divergence in strategies, with China democratizing access to AI tools and the US focusing on consolidation and control.

3. Market Optimism and Uncertain Sustainability: The announcement of the fund has led to positive market reactions, with tech giants like Oracle, Nvidia, and Microsoft seeing stock increases. However, questions remain about the long-term sustainability of these investments and whether the current AI momentum will translate into lasting revenue and innovation.

   Subscribe now

YouTube Episode

Spotify Podcast

Share

Episode Transcript

Opening and Introduction

Chris: Welcome to another episode in 2025 of the Chris Rod Max show. Every week we discuss the latest developments in AI and talk to interesting business leaders and entrepreneurs building the next generation of innovation in this space. I’m excited to be here today with my co-hosts Max and Rod.

Max & Rod: Hello everyone, It's great to be back.

Trump Administration’s AI Fund Announcement

Chris: It’s been an interesting week with the new Trump administration starting their second term. There’s been significant news around AI - specifically about an AI fund. The US is looking to maintain their forefront position in AI technology development with a proposed $500 billion fund to invest in data centers. At the inauguration, we saw representatives from SoftBank, Oracle, and OpenAI present. The initial $100 billion will be deployed immediately into massive data center infrastructure. Trump appears strongly supportive, even removing some AI security regulations. Max, as an investor, what’s your take?

Max: Money is ultimately an incentivizer. Interestingly, this $500 million actually includes Japanese investment, with Masayoshi Son trying to curry favor from an AI project perspective. I see three key points:

1. We’re seeing CEOs changing positions to align with the administration

2. It reflects Trump’s ‘Make America Great Again’ focus - pulling everything back to America

3. The American influence will only grow

If you’re building AI, this signals that your chances of success might be higher if you move to the US. That’s the message President Trump wants to send.

Scale and Controversy

Rod: There’s so much to unpack here. Just two weeks ago, we discussed Microsoft’s $80 billion AI development announcement for this year alone - comparable to some countries’ GDP. Now we have this consortium planning to deploy up to $500 billion over four years. However, there’s controversy because Elon Musk claimed only $10 billion has been secured so far, which sparked debate about his relationship with Trump.

OpenAI’s Role and Market Dynamics

Rod: Looking at AI developments, we see two significant aspects:

1. OpenAI is leading this, despite not being ‘open’ - they’re actually closed

2. They’re securing infrastructure that will further entrench their market position

This makes it harder for new players to enter. Consider Europe - who there can invest $500 billion in their own model development?

Global Competition and Strategy

Chris: Two interesting observations: 1. The consortium includes OpenAI, SoftBank, Oracle, and MGX (Abu Dhabi-based) 2. The political dynamics - Trump’s relationship with tech leaders is evolving

Trump seems to be diversifying his tech industry support, not just relying on traditional allies. Even Mark Zuckerberg attended the inauguration.

Chinese Competition and Innovation

Rod: Let’s discuss DeepSeek, a Chinese company gaining attention in the AI research community. They’re providing their models free of cost, similar to Meta’s LLAMA models. They recently released their R1 model, claiming it surpasses OpenAI’s latest models.

Max: It’s ironic - the US typically democratizes access, but now they’re moving toward centralization. Meanwhile, China is taking the opposite approach, making their technology more open. We’ve seen this with Tencent’s recent release of open-source 3D modeling AI.

Market Impact and Future Outlook

Chris: The market has responded positively - Oracle up 7%, ARM 17%, Nvidia 4%, Microsoft 3%. The business sector seems to view this as a positive signal.

Max: From an investment perspective, we need to consider Warren Buffett’s wisdom: “In the short term, the stock market is a beauty contest; in the long term, it’s a weighing machine.” AI is currently the hot topic, but only time will tell if it’s sustainable from a revenue perspective.

Closing Thoughts

Chris: This has been a special episode covering the announcement of the $500 billion AI fund, endorsed by Trump and starting with a $100 billion investment in data centers. We’ve explored the US-China AI race and China’s emerging open-source platforms. Thank you, Rod and Max, for your insights. To our audience, please subscribe to our newsletter and channels, and we look forward to your feedback.

Your Hosts

Christine Wang Rod Rivera Maxson J.Y. Tee

In this episode, Rod and co-hosts Chris and Max engage with Hatem Diabi, the co-founder and CTO of Inboundr.ai, to explore the evolution of AI agents in content creation.

Subscribe now

Hatem shares his journey from a failed food delivery startup to developing Inboundr.ai, a tool designed to help professionals create authentic content for social media.

The conversation delves into the challenges of maintaining authenticity in online communication, the data sources that inform Inboundr.ai’s AI, and the importance of building a strong team.

Hatem also discusses the integration of Inboundr.ai with Slack, the future of video content, and the pricing strategy that positions Inboundr.ai as a valuable service in the market. In this conversation, Hatem discusses the evolution and impact of AI agents in the realm of inbound marketing.

He emphasizes the importance of expertise in AI solutions, the value proposition for clients, and the simplicity of pricing strategies. The discussion also covers the practical demonstration of Inboundr.ai, a tool designed to enhance social media engagement while maintaining authenticity.

Hatem shares insights for founders in the AI space, stressing the need to focus on solving real problems and the importance of starting early in building a personal brand.

Share

Chapters

• 00:00 Introduction to AI Agents and Inboundr.ai

• 01:45 The Journey from Food Delivery to AI Content Creation

• 03:02 Understanding Inboundr.ai’s Functionality and User Experience

• 09:31 Authenticity in Content Creation

• 10:06 Data Sources and Learning Mechanisms

• 13:23 Building the First Version of Inboundr.ai

• 16:29 Integration with Slack and Future Developments

• 19:12 The Future of Video Content in Marketing

• 22:09 Team Building and Attracting Talent

• 26:22 Pricing Strategy and Market Reception

• 27:20 The Evolution of AI Agents

• 28:08 Client Insights and Value Proposition

• 30:01 Pricing Strategies for AI Solutions

• 31:51 Demonstrating Inboundr.ai: A Practical Walkthrough

• 39:00 Engagement Beyond Posting: Future Features

• 40:55 Maintaining Authenticity in AI-Driven Interactions

• 46:56 Key Takeaways for Founders in AI

• 48:57 Final Thoughts and Call to Action

Takeaways

1. Authenticity Drives Engagement in AI Content Creation

   Inboundr.ai demonstrates that authenticity is a cornerstone of successful content on platforms like LinkedIn. By leveraging AI to capture a user’s unique tone and style, the tool helps professionals craft posts that resonate deeply with their audience. This approach challenges the generic outputs of traditional AI models and highlights the value of genuine, personal storytelling in building a strong personal brand.

2. The Shift Toward “Value as a Service” in AI Pricing

   Inboundr.ai’s pricing strategy underscores the transition from traditional SaaS models to “value as a service.” By focusing on domain expertise, such as LinkedIn content, the platform justifies a premium price point compared to generalist tools like ChatGPT. This evolution reflects a broader industry trend where niche-focused AI solutions are redefining how value is delivered and monetized in the AI market.

3. AI Tools as Catalysts for Personal Branding

   The discussion with Hatem Diabi highlights the growing importance of personal branding for professionals and executives. Inboundr.ai not only simplifies content creation but also helps users build the habit of consistent posting. This positions AI tools not just as productivity enhancers but as essential partners in helping individuals establish and grow their professional presence online.

Subscribe now

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction

Rod: Welcome to another episode of the Chris Rod Max show. Every week, together with AI founders, we discuss what’s happening in the AI space, what they’re doing, and how they’re impacting industries and our society. Before we get started, remember that we have our newsletter on our website, chrisrodmax.com. Subscribe there to get the latest news, updates, and all of our shows.

Today we have as our guest Hatem, who is a co-founder and CTO of Inboundr.ai. We will discuss with him what it means to build a product around AI agents, how he got started, and how he sees the space of AI agents. So, Hatem, welcome. Thank you for being here today.

Hatem: Thanks, Rod. Thanks for having me. I’m Hatem, CTO and co-founder of Inboundr.ai, which is an AI agent for content creation. We help professionals build their personal brand through content that is authentic. And also, we help them build the habit of posting regularly.

Rod: And of course we have our two co-hosts, Chris and Max.

Chris: Hi everyone. Good to be back. Excited about hearing your story, Hatem.

The Origin Story

Rod: So let’s start there. How did it start? Tell us about day one. How did you have this idea of having an agent for content creation in social media? How did that happen?

Hatem: Yeah, so actually, me and my co-founder Sean, we started this after a failure. We had a startup in the food industry - it was like a homemade food delivery app that we had to shut down because it was going viral in the city. The government called us and said we were risking a $20,000 fine. So we had to shut down the startup and build a new one.

We had tons of ideas, and then Sean started talking about inbound marketing. To be honest with you, at that time, I didn’t know what inbound was at all, so he had to teach me. It was the end of 2023, and a lot of people were revising their sales quota. At that time, many people realized they would never reach their quota. The only ones who reached the quota were people who were creating content, especially on LinkedIn. So this is how we started focusing on this problem.

Understanding the Tool’s Value

Rod: So out of necessity - you had to find something new and then turned out to be here in inbound marketing. I find that interesting. Max, Chris, you are not necessarily the most active people on social media. Maybe from your perspective, how do you see these types of tools? Maybe how would you use this type of tools?

Max: So Chris, you’re not super active in social media and then Hatem has Inboundr.ai, right? And with Inboundr.ai, I understand it generates for you just topics and posts that you can just copy paste and get engagement on your profile. Is that correct, Hatem?

Hatem: Yeah, so basically we help you capture the best moments of your podcast or any YouTube URL you have or any Slack discussion you have. Inboundr.ai captures everything and proposes the best things to talk about on LinkedIn, whether it’s success or failure.

If you go to my LinkedIn, yesterday I had a big bug in my software. Inboundr.ai proposed that topic and it was a bug, but it was really an authentic story to share. This is the type of topics that we miss as humans. We have a lot to say, but sometimes we are shy. Sometimes we think that just talking about success is the only way to build the brand. But honestly, people will react to your posts only when you’re authentic, and part of it is sharing failure.

The Technology Behind Authentication

Chris: So essentially what I’m understanding, it’s a tool or product that helps to turn any information bits and pieces into some coherent content that can be reshared. Tell us a little bit more, how do you get to a very natural tone? I mean, if you use some of these foundational models like ChatGPT or whatever, by now, I think everyone can tell whether a text is actually composed by an AI or by a human.

Hatem: Great question. So actually we have our chief innovation officer focusing on how to make content authentic first. Our AI agent goes to your LinkedIn profile, your blogs and podcasts you have participated in. With that, it knows how you talk and what kind of words you use. Are you North American? Are you African? All this plays into the way you express yourself.

Our AI agent captures that regularly. Also, as a human, you evaluate and learn new words, so our AI agent learns with you. Anything you share, it will learn from your tone and apply it to any posts it generates.

Business Model and Pricing Strategy

Chris: Looking at the business model, we can see the pricing starts at $8 per month plus $30 per each user. For some, that might sound a bit steep if we think ChatGPT costs around $20 per month. But also you are saying that some CEOs are paying for ghost writers thousands of dollars. What has been the reception for this specific pricing and how did you come up with this number?

Hatem: Great question actually. When you look at pricing evolution over the last 20 years, it evolved a lot. Right now, we’re seeing a new trend in pricing which some people call “value as a service.” We don’t offer SaaS anymore, we offer value.

Think about ChatGPT as a generalist - someone who knows a little bit about everything but isn’t an expert. You wouldn’t pay someone like that $200 to create your post, maybe $15-20. That’s okay because they’re not an expert. With AI agents, we’re building experts in one domain. My AI agent does one thing - LinkedIn posts - but does it better than any ghostwriter.

Team Building and Growth

Rod: And Hatem, you were also mentioning that you have now an innovation person that is really looking at how this tech can sound as natural and on-voice as possible. Given that you had this idea before for the food app, I imagine you have a team in place. Were you able to move with this existing team and develop this new topic or did you have to hire new profiles? How did you go about the team building process?

Hatem: Great question. So we started with me and Sean. Then we realized we needed an AI expert. I do a lot of things - DevOps, backend, frontend, and everything in between. I also do support. So it’s a lot for me to do everything and do it well.

What we did is bring in two other co-founders. They’re both ex-Microsoft. One is MG, who has the best course on machine learning on Udemy. If you go to Udemy and type ‘machine learning,’ he’ll be the first guy. He has like 20,000 students. And also he’s a YouTuber with thousands of followers. That guy is our innovation officer.

The other co-founder is Samir who is a professor at McMaster and also one of the top 200 employees at Microsoft worldwide. We brought the best talent from Microsoft in the AI field.

Future of Social Media Interaction

Chris: And let’s imagine that Inboundr.ai is very successful. So every post coming on LinkedIn was generated with Inboundr.ai and even replied to. Where would be the space for human interaction? Will there still be human interaction on LinkedIn or are we approaching a future where it’s only agents talking to each other and then humans are somewhere else doing something different?

Hatem: Yeah, I think we don’t want to reach that point. The whole reason for doing this is to make LinkedIn authentic because probably when you open your feed, there are a lot of ChatGPT posts there. People don’t even know what they’re talking about - it was just proposed by ChatGPT, so they post.

We want to reach a point where you really build connections with people. This is why you want to build your personal brand. This is just a way to do it faster, not about becoming an influencer and forgetting about your responsibilities as a C-level executive or as a developer.

Posting is a way to get leads, but then from that moment it’s your responsibility to take that lead and build the network and communication.

Closing Thoughts and Contact Information

Rod: So there it is for everyone, right? Go to Inboundr.ai, sign up, get onboarded. I actually had my onboarding yesterday, so I’m looking forward to starting using it.

Hatem: Thanks Rod, Chris, Max. It was a pleasure. Let me know when you want to start. And last advice for everyone - start as soon as possible. I started late and I regret it. Every time I meet with people, I remind myself I built this for you - for people who are busy, who work hard, but don’t have time to build a personal brand. So anytime there’s a hard worker who starts posting, I get very happy. Please reach out if you need some help.

Rod: Very motivational! So everyone, don’t overthink it, just do it. See you everyone!

Your Hosts

Christine Wang Rod Rivera Maxson J.Y. Tee

In this episode, Max and Rod discuss the latest developments in AI, focusing on Microsoft’s significant investment in AI infrastructure, the implications of US-China relations in the AI landscape, and the future of AI in the workforce.

Subscribe now

They reflect on Sam Altman’s insights regarding OpenAI’s growth and the potential for AI agents to transform work dynamics. The conversation emphasizes the need for training in AI usage and the evolving expectations in the workplace as AI becomes more integrated into daily tasks.

In this conversation, Max and Rod explore the current landscape of AI investment, discussing the staggering amounts of funding flowing into AI startups and the implications for the market. They delve into the evolution of AI companies, the potential for personal AI assistants, and advancements in AI reasoning models.

The discussion also covers the impact of AI on content creation, including the rise of AI-generated media and podcasts. Finally, they analyze NVIDIA’s strategic move to open source its recent acquisition, considering the broader implications for the AI community.

Chapters

• 00:00 Welcome Back and Reflections

• 01:13 Microsoft’s Massive AI Investment

• 05:03 The US-China AI Landscape

• 08:39 Training for an AI-Driven Future

• 12:41 The Role of AI in Daily Life

• 15:16 Evaluating AI Investments

• 17:23 Sam Altman’s Reflections on OpenAI

• 21:45 The Future of AI Agents

• 25:51 Adapting to AI in the Workforce

• 28:32 The Rise of AI Investment

• 30:11 Funding Trends and Market Dynamics

• 32:56 The Evolution of AI Companies

• 36:36 The Future of Personal AI Assistants

• 39:18 Advancements in AI Reasoning Models

• 41:43 The Impact of AI on Content Creation

• 44:19 AI-Generated Content: The Future of Media

• 51:53 NVIDIA’s Strategic Moves in AI

Takeaways

1. Microsoft's $80 Billion AI Infrastructure Investment by 2025

Microsoft’s announcement to invest a staggering $80 billion in AI infrastructure by 2025 showcases the immense scale of commitment required to stay competitive in the AI industry. This figure is more than the GDP of several countries, emphasizing the transformative role AI is expected to play in global economies. The investment highlights Microsoft's belief in AI’s potential to redefine industries, create new job markets, and fuel technological breakthroughs. It also reinforces their strategic partnership with OpenAI, further positioning Microsoft as a leader in the AI space.

This level of investment not only accelerates innovation but also signals to other players in the tech industry that the race for AI dominance is becoming increasingly capital-intensive. It underscores the need for both public and private sectors to collaborate on infrastructure, talent development, and ethical AI frameworks to ensure sustainable growth.

2. The Potential for AI Agents to Join the Workforce by 2025

The concept of AI agents joining the workforce by 2025 is one of the most exciting developments discussed. These agents, powered by advanced reasoning models and autonomous decision-making capabilities, have the potential to revolutionize productivity in the workplace. AI agents could handle repetitive tasks, optimize workflows, and even collaborate with human workers in creative and complex problem-solving scenarios.

This development reflects a broader trend toward “agentic AI,” where systems are designed to independently perform tasks across domains. As organizations prepare for this shift, reskilling and upskilling the workforce will be essential to integrate AI agents effectively. While the timeline is ambitious, it aligns with rapid advancements in generative AI, reasoning models, and real-time decision-making systems. The introduction of these agents could mark the beginning of a new era in workplace efficiency and innovation.

3. NVIDIA’s Open-Source Strategy and Industry Influence

NVIDIA’s decision to open-source software from their $700 million Run AI acquisition represents a significant shift in their strategy. As the de facto leader in AI hardware, NVIDIA is already indispensable to the industry. By open-sourcing key technologies, they are likely aiming to strengthen relationships with developers and the broader AI community, fostering innovation and goodwill.

This move is also a reflection of NVIDIA’s confidence in their ecosystem. By allowing broader access to their tools, they encourage adoption of their hardware while building a loyal community. However, the risk of diminished ROI on acquisitions remains a concern. Despite this, the potential long-term benefits—such as increased market influence and industry standardization—may far outweigh the costs. NVIDIA’s open-source initiative could set a precedent for other tech giants, further blurring the lines between competition and collaboration in the AI landscape.

Sources

1. The Golden Opportunity of American AI by Brad Smith, MSFT President- https://blogs.microsoft.com/on-the-issues/2025/01/03/the-golden-opportunity-for-american-ai/

2. Reflections - Sam Altman https://blog.samaltman.com/reflections

3. Sapphire’s Top 10 AI Trends & Predictions - https://sapphireventures.com/blog/top-10-ai-trends-predictions-for-2025-a-platform-shift-in-the-making/

4. Top Tech Enterprise Tech experts and their AI views - https://www.ai21.com/blog/2025-predictions-for-enterprise-ai

5. NVIDIA to OpenSource Run.AI - https://venturebeat.com/ai/nvidia-acquires-software-maker-runai-to-orchestrate-gpu-clouds-for-ai/?utm_source=tldrai

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Welcome Back

Max: Welcome to another episode of the Chris Rod Max show. Today we have Rod and Max here.

Max: It’s good to be back. Feels like I missed the world by being gone for a couple of weeks. But yeah, things moved.

Max: How are you feeling today, Rod?

Rod: I’m feeling great and, exactly as you’re saying in our previous episode, people ask me, where is Max? What happened to him? Where were you, Max?

Max: Yeah, good question. So for those who are listening, I was actually very ill. Unfortunately, I was traveling to Asia, but I somehow got food poisoning, which is not great. But yeah, I mean, nevertheless, I am fully recovered now, you know, new year, new you kind of thing. And it’s good to be back. I’ve missed doing this, to be honest.

Newsletter Announcement

Rod: New Year New You and yeah, of course. That’s the thing to do now. If everyone did not know yet, we have a new newsletter, which you should subscribe to, where we try to put on all the latest and the greatest news on AI for business up there. So remember to subscribe.

Microsoft’s AI Investment Discussion

Max: So today, given that we’re all coming back from holiday, there’s actually quite a lot of action going on for the last two weeks or the first two weeks of the year. And we’re going to spend a little bit of time talking about Brad Smith, the Microsoft President’s latest article. And then we’re obviously going to spend a little bit of time talking about Sam Altman’s reflection of what he has done for last year, which was pretty eventful from an OpenAI perspective.

Rod: When we were talking about this, we spent some time talking about the biggest protagonist on AI, which has been Microsoft and OpenAI, obviously when they invest such a huge amount. And in there, we also have Brad Smith talking about their plan to invest about $80 billion in AI infrastructure for FY 2025. That’s eight-zero, guys. That’s eight-zero. That’s a lot of money.

Max: For just the number, I was thinking about that and you think about this: this is more than the GDP of Sri Lanka or other countries such as Azerbaijan, Tanzania, or about the level of Serbia. So in one year, Microsoft is investing the same amount of money that a whole country produces for its GDP.

The US-China AI Competition

Rod: Yes, and this division, US-China, is really a common theme in our show. So we discuss all the things that are happening in China that very often we are not really aware of. We’re seeing how through legislation, through business decisions, and so on, we’re seeing a rift between what is happening in the Western world with companies such as Microsoft here, but also in China with all the local players.

AI Training and Implementation

Max: So allegedly, one billion future AI jobs and so on. So of course, seeing from that perspective, we need $80 billion per year in investment in AI to have the infrastructure in place and everything necessary to foster this innovation and these future jobs.

Rod: If we think back to the late nineties and so on, there used to be internet cafes. For example, Apple stores were places where people would come for training: “How can I send emails? How can I browse the internet?” Now, of course, this is almost intuitive. Nobody really needs a boot camp or training course anymore.

OpenAI’s Growth and Success

Max: Speaking of looking hard enough, Sam Altman obviously recently published an article on his own blog reflecting on what happened last year. For those who don’t follow this very closely, Sam Altman was ousted and then reinstated as CEO, then a couple of co-founders left. But in the middle of all this, OpenAI was still growing quite significantly.

Rod: It’s estimated that ChatGPT is now generating $2.7 billion every year from subscriptions. People are paying $20 per month. I mean, how many companies can claim they generate this amount of money? It’s astronomical, and this is something that two years ago, three years ago didn’t exist at all.

Future of AI and Personal Assistants

Max: I think in the article, Sam Altman actually made the bold claim that they now know—or are confident—about how to build an AGI. He foresees that in 2025, AI agents will join the workforce.

Rod: Certainly, Max. I saw this quote about AI agents in 2025. Some estimate that we’ll see the first AI agents. Last week with Chris, we were also saying that this year will be the year of the agent.

NVIDIA’s Strategic Moves

Max: In the interest of time, we have one last segment, which is about NVIDIA. I want to talk a little bit about this. Just interesting to get your take. We all know now NVIDIA is the darling of AI. Everybody wants their chip. And we also know that NVIDIA bought Run AI for $700 million. And now they’ve decided to open source the software they bought with Run AI.

Rod: This acquisition is so irrelevant in the bigger picture for NVIDIA that even if it doesn’t pan out and even if it’s not a good idea to offer this for free instead of commercializing it, that doesn’t really change anything for NVIDIA.

Closing Remarks

Max: With that, we are on top of the hour, slightly over. I think we really enjoyed the chat. If you like what you’re hearing, share it with your friends, comment, or like the post. We really appreciate any feedback that you have. It is only your feedback—your listening—that can help us improve the podcast. And with that, thank you so much again for joining us for another episode and happy 2025. May this year be the year of agents and whatever it is that you have going on.

Rod: Thank you very much and tune in for the next one. And don’t forget to subscribe to our newsletter on chrisrodmax.com.

Max: Good one, Rod. Subscribe. Do what Rod says.

Your Hosts

Christine Wang Rod Rivera Maxson J.Y. Tee

In this episode of the Chris Rod Max Show, hosts Chris and Rod discuss the latest developments in AI, focusing on market predictions for 2025, the role of NVIDIA, the impact of big tech companies, energy consumption concerns, data constraints, and the future of startups in the AI space.

They explore the potential for market consolidation and the evolving landscape of AI technology, emphasizing the importance of innovation and sustainability. In this conversation, Chris and Rod explore the evolving landscape of AI, focusing on corporate markets, startup trends, and the rise of AI agents.

They discuss the implications of AI in various sectors, the consumer perspective on AI technology, and make predictions for the future of AI, including its integration into enterprise solutions and robotics.

Chapters

• 00:00 Welcome to the AI Revolution

• 02:38 NVIDIA’s Market Position and Future Predictions

• 07:17 The Role of Big Tech in AI Development

• 10:08 Energy Consumption and Environmental Impact of AI

• 17:44 Data Constraints and the Future of AI Models

• 26:07 Market Consolidation in the AI Space

• 28:43 The Future of Startups in AI

• 29:07 AI Solutions in Corporate Markets

• 30:16 Trends in Startup AI Development

• 32:54 The Rise of AI Agents

• 34:37 Understanding AI Agents

• 38:58 Consumer Perspectives on AI

• 48:58 Predictions for AI in 2024

Takeaways

• NVIDIA's Dominance Faces Challenges: While NVIDIA maintains a strong monopoly in the GPU market, competition from specialized chip providers (e.g., Groq, SambaNova, and Cerebras) and big tech companies like Apple developing in-house hardware could disrupt their dominance. Companies are increasingly seeking cost-efficient and tailored solutions, which might lead to diversified hardware adoption in the AI space.

• Energy Consumption and Sustainability Are Growing Concerns: As AI models consume more energy, sustainability is becoming a critical factor for businesses. This could lead to a shift toward on-device solutions and push companies to prioritize greener AI technologies to align with ESG metrics and consumer sentiment.

• The Rise of AI Agents: AI agents are poised to become ubiquitous in 2025, especially in B2B applications where they can streamline operations like invoice processing. However, consumer adoption may lag due to trust and usability challenges. The integration of agentic functionality across SaaS products has the potential to revolutionize processes and deliver significant value across industries.

Subscribe now

YouTube Episode

Spotify Podcast

Episode Transcript

Rod (00:01.564)

Welcome to another episode of the Chris Rod Max Show, our first episode of the year! Hi Chris, how are you doing today?

Chris (00:10.654)

Fantastic! Happy New Year to everyone! I hope you all had a great celebration as we welcomed 2025.

Rod (00:20.402)

Great to hear that, Chris! For those tuning in for the first time, this is the show where we dive into the latest developments in AI every week. We discuss how businesses can implement AI to improve efficiency and achieve better outcomes.

Today, we’re tackling what 2025 holds for the world of AI. We’ve prepared a packed agenda exploring predictions from experts in the space and adding our own perspectives. The big question is: are we nearing the end of the GNI revolution, or is it just getting started?

Let’s begin with a topic we covered extensively last year—NVIDIA. It feels like everyone knows NVIDIA now. The constant debate on our show has been about their dominance in the GPU market. Will their monopoly hold, or will competitors start to disrupt the space?

Predictions from Vivek Ramaswami and Sabrina Wu suggest 2025 could be a turning point. They foresee specialized chip providers like Groq, SambaNova, and Cerebras gaining traction, possibly cutting into NVIDIA’s market share. What’s your take, Chris?

Chris (02:38.06)

That’s a great question, Rod, and one many investors are asking. NVIDIA’s stock skyrocketed last year, unlike other semiconductor companies like Micron, AMD, and Intel, which struggled.

NVIDIA still enjoys a monopoly premium, but history tells us competition eventually catches up. Additionally, companies in the AI space are exploring in-house chip development to reduce reliance on NVIDIA. What are your thoughts?

Rod (03:56.082)

I agree. While NVIDIA remains dominant, reports suggest competitors like Intel are catching up in performance. However, NVIDIA’s ecosystem is their strongest asset. They’ve built a developer-friendly platform that’s hard for others to replicate.

For instance, I spoke with someone comparing the costs of deploying voice AI applications on different platforms. Using Groq was significantly cheaper—around $10 per month compared to $200 with OpenAI. This massive cost difference could make companies reconsider their options.

Chris (06:49.218)

Absolutely. Besides specialized chip providers, we also have major tech companies like Apple entering the space. Apple’s control over their supply chain and potential GPU development could reduce reliance on NVIDIA. How do you see this playing out?

Rod (08:14.77)

I see a clear segmentation. Consumers may stick with OpenAI or Apple’s AI features, while SMBs and enterprises could explore alternatives like Groq for cost efficiency. Meanwhile, cloud providers and model developers like OpenAI might move toward building their own chips to optimize for specific architectures.

Rod (10:08.588)

Energy consumption is another critical issue. With AI models consuming increasing power, there’s potential for a shift toward on-device solutions. Similar to how crypto faced backlash for its environmental impact, AI providers could face scrutiny. Do you think sustainability will influence consumer sentiment or regulatory action?

Chris (11:41.096)

That’s a nuanced issue. Consumers are more likely to care about data privacy than energy consumption in AI. However, businesses might prioritize sustainability if ESG metrics and investor demands push them to choose greener AI solutions.

Rod (16:14.822)

Speaking of constraints, Ilya Sutskever predicts a fundamental shift in how we train AI models due to data scarcity. Public data may no longer suffice, and synthetic or proprietary data will become critical. Chris, do you think we’re approaching “peak data”?

Chris (17:44.942)

Yes, public data is reaching its limits. Synthetic and proprietary data are the way forward. But creating genuinely novel data is challenging, and models may face diminishing returns without truly diverse datasets. What do you think, Rod?

Rod (19:39.506)

Exactly. The novelty of data is crucial. For instance, how many pictures of the Eiffel Tower do we really need? Companies must innovate in data generation to overcome this bottleneck.

Rod (27:16.718)

On startups, there’s speculation about market consolidation in 2025. Many AI companies funded in the past few years now need new rounds, but with large players flush with cash, we might see more acquisitions. Chris, what trends are you noticing in the startup world?

Chris (30:41.294)

I see three categories: rebranded companies adding “AI” to their pitch, middleware/infrastructure builders, and foundational model developers like Mistral. The latter is a high-stakes game, requiring significant investment and expertise.

Rod (34:37.378)

That brings us to AI agents, which John K. Thompson predicts will become ubiquitous in 2025. While Salesforce and others have started experimenting, we’ve yet to see true agentic solutions scale widely. Chris, will this be the year of AI agents?

Chris (36:13.424)

In B2B, yes. AI agents can streamline operations like invoice processing. In B2C, adoption might lag due to trust and usability challenges. The key is overcoming early hurdles to deliver meaningful value.

Rod (38:40.592)

I believe AI agents will become essential, enabling models to interact with their environments. Every SaaS product could incorporate agentic functionality, revolutionizing processes across industries.

Chris (41:03.146)

That’s a compelling vision. But as we’ve seen, even basic AI features like adding events to calendars still face technical limitations. There’s room for improvement, especially in consumer-facing applications.

Rod (50:26.962)

Let’s wrap up with our predictions for 2025. Mine are: the rise of AI agents, AI’s growing political influence, and AI-generated video becoming mainstream. Chris, what are yours?

Chris (51:43.938)

I predict: enterprise AI adoption delivering measurable ROI, AI breakthroughs in life sciences, and the convergence of AI with robotics for real-world applications.

Rod (54:34.978)

Great insights, Chris! Let’s revisit these predictions mid-year to see how they hold up. Thanks to our listeners for joining us! Remember to visit chrisrodmax.com for more episodes and subscribe for updates. Until next time!

Your Hosts

Christine Wang Rod Rivera Maxson J.Y. Tee

Subscribe now

In this conversation, Anna Spitznagel discusses the importance of automating AI governance and compliance processes. She explains how her company, Trail, provides a co-pilot for AI governance, helping organizations navigate the complexities of AI regulations. The discussion covers practical examples of implementing AI solutions, customer onboarding, pricing models, and the significance of building trust with clients. Anna also shares insights into the future of AI governance, emphasizing the need for clarity in regulations and fostering trust in AI technologies.

Subscribe now

Takeaways

• The EU AI Act is the first large-scale regulation of AI globally.
  

• AI governance is crucial to align AI technology with European values.
  

• Companies must categorize their AI use cases into risk classes.
  

The EU AI Act is the first large-scale regulation of AI globally.

AI governance is crucial to align AI technology with European values.

Companies must categorize their AI use cases into risk classes.

Like what you hear? Remember to smash that subscribe button for more insights every week!

Find below👇 our:

• YouTube Episode

• Spotify Podcast

Subscribe now

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Welcome

Chris: Welcome to another episode of the Chris Rod Max Show! I'm really excited to have Rod and our special guest Anna from TrailML joining us today. Even though it's almost the end of the year, we're fortunate to have Anna in our virtual studio to discuss the EU AI Act, governance, and data security.

Rod: Hello everyone!

Chris: How are you both doing today? Rod?

Rod: I'm really happy we can close out the year with Anna, very glad to be here.

Chris: How about you, Anna?

Anna: I'm delighted to be here! This is actually my second podcast, so I'm becoming quite the pro. Looking forward to discussing responsible AI and AI governance with you both.

Understanding the EU AI Act

Chris: Before we dive into what you all do, it would be really helpful for our audience to understand exactly what the EU AI Act is. What was the idea behind it? Where do we stand now? And how are companies and stakeholders adopting it? Anna, could you help us get started?

Anna: Of course! Let me give you a quick overview of the AI Act and its most important aspects. The European AI Act is the first comprehensive AI regulation globally. While there have been regulatory attempts in different parts of the world, the AI Act is the first major regulation that's already been put into force.

The EU chose to regulate use cases rather than the technology itself, which is why they introduced risk classes and roles. These are the most important concepts of the AI Act because they determine what companies need to do when using, developing, or selling AI products.

There are four risk classes:- Prohibited AI- High-risk AI- Limited risk AI- No risk

And for roles, the most common ones I see in companies are:- Provider of an AI system- Deployer of an AI system

Timeline and Compliance

Anna: The AI Act was put into force in August this year, and the first deadline is coming up in February 2025. This initial deadline has two key components:

1. Prohibited AI systems must be removed from the market

2. Organizations must ensure AI literacy across their workforce

Regarding enforcement, the fines are substantial - up to €35 million or 7% of global revenue, whichever is higher. This was intentionally set high by the EU.

Enforcement and Implementation

Chris: That's quite comprehensive. If I can recap: we have four risk levels, two roles (provider or user), and based on these combinations, there are different requirements to follow. This year marks the first year the EU Act is in force, with specific deadlines and significant fines for non-compliance. Who actually enforces these requirements? Is there someone who checks your systems?

Anna: That's still being figured out. Member states have until August 2025 to determine who will be responsible for enforcement at the national level. In Germany, it looks like it will be the Bundesnetzagentur, but this can vary by country.

For generative AI and general-purpose AI, the EU has established a dedicated AI office at the European level. They're actively hiring, especially technical experts, as reviewing these models requires specific expertise.

Risk Categories and Examples

Chris: Could you explain the four risk categories and provide some examples? Also, tell us more about AI literacy - what does that really mean?

Anna: Let me break down the risk categories:

1. Prohibited AI: These are use cases the EU won't allow on the market. For example, social scoring by governments or emotional recognition in workplaces.

2. High-Risk AI: This typically involves human data or decisions that impact people's lives. A good example is credit scoring - AI systems that determine whether someone gets a loan. These systems are allowed but must meet strict requirements around risk management, quality management, and technical transparency.

3. Limited Risk AI: This mostly applies to generative AI solutions that interact with people. The focus is on transparency - ensuring people know when they're interacting with AI or when content is AI-generated.

4. No Risk: Simple applications like email spam filters, where the potential harm is minimal.

AI Literacy Requirements

Anna: Regarding AI literacy, it's about ensuring every employee who works with AI understands its basics. The AI office recently clarified that everyone needs a base understanding that AI isn't a magic box but uses mathematical methods. They need to understand AI's limitations and be aware of the AI Act's rules.

Companies are approaching this through various methods, often starting with webinars for basic training. The level of required knowledge varies by role - data scientists need deeper understanding than business users.

Industry Impact and New Roles

Chris: Sometimes governments and companies might not align because it's one thing to create policies and another to upskill people. Have you seen new types of jobs emerge because of the EU Act and the rise of generative AI?

Anna: Yes, we're seeing new roles emerge. Companies are creating positions like AI governance leads, which were previously luxury positions only at big companies. The question of who owns AI governance and compliance varies - sometimes it's in IT security, sometimes in compliance, and sometimes under the AI manager.

TrailML's Role

Chris: Tell us more about TrailML.

Anna: TrailML provides tools to help with AI regulation compliance. Many companies initially think they can manage with Excel, but AI's dynamic nature requires more sophisticated solutions. We built Trail as a co-pilot for AI governance, automating documentation and ensuring transparency.

We're currently a team of 10 people, and we've developed a co-pilot that guides organizations through compliance requirements. Our system integrates with code and databases to provide structured information, making governance processes more efficient.

Practical Application

Chris: Let's use an example. Say Rod and I have the Chris Rod Max Show podcast company and want to implement an AI chatbot. What happens when we come to you for compliance help?

Anna: You would start by logging into our web platform. If you're developing the chatbot yourself, you'd use our Python package to upload your code and logging information. The platform would guide you through categorizing your use case and determining your risk class and role.

In the case of a chatbot, it would likely fall under limited risk, which means fewer requirements than high-risk applications. You'd assign roles and responsibilities - perhaps one compliance lead, one AI lead, and one developer. Each person would handle different requirements and approvals.

Customer Base and Pricing

Chris: Who are your customers? Are you targeting smaller or larger companies?

Anna: In theory, everyone needs AI governance, but we currently see two main customer types:

• Large enterprises who have been thinking about this for a while and need smart solutions for their complex regulatory environment.

• AI-native startups who need to prove compliance to sell to larger companies.

Our pricing is use-case based - it differs significantly between a single use case and handling 100 use cases. For smaller companies, think of it in the productivity tool price range; for enterprises, it's more in line with typical compliance solution pricing.

Looking Ahead

Rod: As we approach 2025, what are your predictions for next year? Where do you see the industry going?

Anna: I think we're moving from experimentation to scaling AI solutions and ensuring business value. On the regulatory side, we'll see more clarity through standards, guidelines, and best practices. For general AI, I expect continued innovation, fewer hallucinations, and better quality models.

Final Thoughts

Chris: Anna, for any listeners wanting to get in touch with you, how can they do that?

Anna: You can check out the TrailML website, book a demo, or reach out to me directly on LinkedIn. Happy to chat about all these topics!

Chris: Thank you for sharing your knowledge about the EU AI Act. I'm sure our listeners learned a lot today.

Rod: Don't forget to subscribe to our newsletter on ChrisRodMax.com!

Chris: Thank you, Rod.

Anna: Have a great day!

In this episode, Rod discusses the most significant developments in AI for 2024, including the awarding of Nobel Prizes for advancements in neural networks, the rise of AI agents, the explosion of multimodal AI capabilities, the establishment of regulatory frameworks, the growth of open-source models, and the rapid adoption of AI technologies in enterprises and creative industries. The conversation highlights both the positive impacts and challenges posed by AI on job markets and society as a whole.

Takeaways

• The Nobel Prize in physics was awarded for neural networks.

• Generative AI is now a key player in creative industries.

• AI agents are enhancing productivity across sectors.
  

Like what you hear? Remember to smash that subscribe button for more insights every week!

Find below👇 our:

• YouTube Episode

• Spotify Podcast

• Transcript

LinkedIn profiles of the hosts

Christine Wang, Rod Rivera, Maxson J.Y. Tee

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction

Rod: Welcome to another episode of the Chris Rod Max show, where we weekly explore the latest developments in AI - from enterprise adoption to consumer technology. This is a special episode as I'm your sole host today. Chris and Max couldn't join us, but that won't stop us from delivering valuable insights.

As we approach the end of the year and holiday gatherings begin, AI discussions are becoming increasingly common. To help you stay informed and contribute meaningfully to these conversations, I've compiled the top 10 most significant developments in the generative AI space for 2024.

1. Physics Nobel Prize Recognition

Rod: Let's start with perhaps the most significant development of 2024. The Nobel Prize in Physics was awarded to Geoffrey Hinton and John Hopfield for their groundbreaking work in neural networks - the fundamental technology behind generative AI. What makes this particularly remarkable is that Hinton had already received the Alan Turing Award - often called the Nobel Prize of computing - back in 2018.

There's an interesting anecdote here. When New York Times reporter Kate Metz asked Hinton to explain the Boltzmann machine in simple terms, he cleverly quoted Richard Feynman, saying, "Listen buddy, if I could explain it in a couple of minutes, it wouldn't be worth the Nobel Prize."

2. Chemistry Nobel Prize Breakthrough

Rod: The second major development is equally impressive - another Nobel Prize, this time in Chemistry, was awarded to DeepMind's team for their work in protein sequence analysis. For context, DeepMind, acquired by Google in 2014, has been pursuing general artificial intelligence with a focus on life sciences.

Their breakthrough was realizing that, just as ChatGPT predicts the next word in a sentence, we can predict protein chains. This development has huge implications for medicine, enabling faster and more cost-effective development of new treatments and therapies. The impact is already visible, with the team behind this technology securing major deals with pharmaceutical companies like Lilly.

3. Rise of AI Agents

Rod: A recurring theme on our show has been the rise of AI agents, and 2024 has seen this trend accelerate dramatically. The SaaS industry has widely embraced AI agents, integrating them into everyday tools like Google Drive and Microsoft's suite. The vision is clear: eventually, we might have AI employees capable of handling complete projects end-to-end.

While we're not quite there yet, AI agents have already significantly boosted productivity by automating manual tasks and making AI more accessible through integration with common web applications.

4. Explosion of Multimodality

Rod: This year marked a significant shift from pure chatbots to truly multimodal AI. While image generation through tools like DALL-E and Midjourney has been around since 2022, 2024 was the first year where we saw widespread access to video and sound generation through simple text prompts.

Take Sora from OpenAI as an example - it can generate various styles of content, from 3D to realistic to cartoonish, all through simple text prompts. What previously required a full production crew can now be accomplished through AI. While these tools aren't perfect, they represent a massive leap forward from just a year ago.

5. Rapid Regulatory Framework Development

Rod: Unlike previous technologies where regulation took years to develop, AI has seen unprecedented speed in establishing regulatory frameworks. This rapid response stems from demand across businesses, governments, and society for clear guidelines. We've witnessed the creation of committees both nationally and internationally, working to establish comprehensive AI regulatory frameworks - a trend that will continue into 2025.

6. Open Source Model Revolution

Rod: Last year, it seemed like AI access would be controlled by a few major players like OpenAI, Anthropic, and Microsoft. However, 2024 brought a dramatic shift. Not only have prices decreased, but open-source models have improved so significantly that they often match their closed-source counterparts.

Models like Mistral, Llama, and Qwen have become extremely reliable and are being adopted by companies worldwide. This could lead to a future where even major players like OpenAI might control only 10-20% of the market due to the abundance of high-quality alternatives.

7. Enterprise AI Adoption Surge

Rod: The enterprise adoption of AI has seen explosive growth. According to McKinsey's April report, AI adoption in enterprises jumped from 56% in 2023 to 72% in 2024 - a dramatic increase compared to the relatively flat adoption rates since 2019. We've moved beyond proof of concepts and pilots to full-scale implementation across industries.

8. Consumer AI Integration

Rod: This enterprise adoption has cascaded down to consumers. Apple Intelligence's public release exemplifies how AI is becoming seamlessly integrated into everyday tasks. Instead of standalone AI applications, we're seeing smart features subtly integrated into existing workflows, from copy-paste functions to background removal and intelligent notifications.

9. Creative Industry Transformation

Rod: The creative industry has been revolutionized by AI. Coca-Cola made headlines with their entirely AI-generated advertisement, and we're seeing similar developments across YouTube, with AI-generated movie trailers and B-roll footage. This extends beyond video into music and art, fundamentally changing how creative content is produced.

10. Real-World Impact and Job Market Effects

Rod: Finally, we're seeing tangible results from AI adoption across industries. Klarna stands out as a prime example, having replaced entire departments with AI systems. They've automated marketing campaigns and customer support, with their AI handling workloads equivalent to 700 support agents. They even used AI-generated voice for their CEO's financial reporting.

However, this transformation has its challenges. We're seeing pressure on salaries and opportunities in creative industries, particularly in areas like copywriting and translation, where project values in the freelance market have significantly decreased.

Conclusion

Rod: These developments show both the impressive progress and complex implications of AI integration across society. Before we end, I encourage you to share your thoughts on these trends in the comments and sign up for our newsletter at chrisrodmax.com for more insights and weekly updates.

Tune in next Monday for another episode with our regular crew. Until then, remember to like, subscribe, and share your thoughts - both positive and negative. Thank you for joining this special episode of the Chris Rod Max Show!

In this episode, Rod and Chris engage with AI expert Andrea Isoni to discuss the latest trends in generative AI as outlined in his annual report.

They explore the distinctions between secular and catalyst trends, the impact of regulation on AI adoption, and the investment landscape, particularly in cybersecurity. Andrea shares insights from his background in data science and the challenges of data generation, emphasizing the importance of both original and synthetic data.

The conversation also touches on the increasing adoption of AI technologies among youth and the implications for businesses. The conversation delves into the evolving landscape of AI usage globally, highlighting significant traffic from countries like India and Indonesia.

Subscribe now

It discusses the trend of companies rebranding as AI-first, the emergence of smaller AI models, and the integration of RPA with AI. The risks of cannibalization in the AI market, the importance of liability in AI development, and predictions for GPU efficiency and energy consumption are also explored, culminating in insights about the future of AI technology.

Takeaways

• The report identifies cybersecurity as a key area benefiting from AI adoption.

• Understanding secular and catalyst trends is crucial for predicting AI's future impact.

• Regulatory frameworks are evolving to protect human life while enabling AI innovation.

Like what you hear? Remember to smash that subscribe button for more insights every week!

Find below👇 our:

• YouTube Episode

• Spotify Podcast

• Transcript

LinkedIn profiles of the hosts

Christine Wang, Rod Rivera, Maxson J.Y. Tee

YouTube Episode

Spotify Podcast

Episode Transcript

Introduction and Background

Rod: Welcome to another fantastic episode of the Chris Rod Max show. I'm joined by my co-hosts Chris and Max and our special guest Andrea Izzoni, an AI expert and industry observer who produces an annual AI report.

Chris: Hey everyone, good to be back.

Andrea: Hi everyone, thank you for having me.

The Structure of the AI Report

Rod: Andrea, you have your yearly report on generative AI, with the last one released in January 2024. Can you share the biggest trends you were seeing?

Andrea: Let me first clarify the report's structure. There are three main sections:

1. Secular trends - trends impacting for 10+ years

2. Catalyst trends - trends impacting for 5 years or less

For example, population aging is a secular trend - it's ongoing and continuous. An example of a catalyst trend would be AI investments, happening for less than five years.

Key Findings and Cybersecurity Focus

Andrea: One significant outcome was the importance of cybersecurity. First adopters of AI are usually malicious actors, and if there's one winner in AI adoption, it's cybersecurity. The more AI is adopted, the more you need to spend or regulate to combat misuse.

In consumer markets, we predicted ChatGPT adoption rates, which proved accurate. We also predicted that smaller companies and open-source models would catch up with proprietary models like GPT-4, which happened around July-August.

Andrea's Background in AI

Rod: Your report is very comprehensive. Can you share your background?

Andrea: I'm an academic translator for business. After my PhD in London in 2010, I moved into data science when it was still a new field. In 2014, many recruiters didn't even know what data science was - at least in London, though it was already established in the US.

I landed a job at Founders Factory, wrote a book called "Machine Learning for the Web," and worked on various AI technologies through consultancy.

Regulatory Evolution and Investment Landscape

Chris: You mentioned cybersecurity getting the better half of AI. Could you elaborate on data privacy and regulation, especially with GDPR and the EU Act?

Andrea: First, let's distinguish between approaches. The EU Act, coming into force on August 1st this year, is a horizontal approach - one body of law regulating all industries. Other governments like Israel or the UK prefer a vertical approach, where each sector's authority handles its own AI regulation.

The EU Act's main principle is protecting human life from AI's unintended effects. It has different risk tiers:- Limited risk (e.g., customer service chatbots)- High risk (e.g., AI scoring examination results)

Future Trends and Hardware Development

Andrea: One clear trend we're seeing involves GPU development, particularly Nvidia's upcoming Blackwell GPU. We're expecting massive performance improvements and significantly lower power consumption. The energy efficiency is improving by 20-30% with each generation.

The hardware landscape shows:- NVIDIA maintaining dominance in corporate/data center space- AMD having opportunities for cost-effective solutions.

Closing Remarks

Rod: Andrea, where can people find your report and connect with you?

Andrea: You can find me on LinkedIn as Andrea Isoni. The report is available for free at stateofartificialintelligence.com or https://www.aitechnologies.co in the AI insights section.

Rod: Thank you everyone for tuning in. Remember to join us every Monday for another episode of the Chris Rod Max Show.

Subscribe now