Pod Insights The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch 20VC: Scale, Surge, Turing, Mercor: Who Wins & Who Loses in Data Labelling | Is Revenue in Data Labelling Real or GMV? | Why 99% of Knowledge Work Will Go and What Happens Then? | Why SaaS is Dead in a World of AI with Jonathan Siddharth @ Turing About this episode Jonathan Siddharth is Founder and CEO of Turing, one of the fastest-growing AI companies advancing frontier models. Jonathan has led the company to an astonishing $300M ARR with just $225M raised and a profitable company. A Stanford-trained AI scientist, Jonathan previously helped pioneer natural language search at Powerset, which was acquired by Microsoft. AGENDA: 03:35 Data, Compute, Algorithms: What is Most Abundant? What is Lacking Most? 09:18 What Does No One Know About AI's Data Requirements That Everyone Should? 17:05 The Biggest Challenges Enterprises Have with AI Adoption 20:38 Why Will 99% of Knowledge Work Will be Gone in 10 Years 27:12 How Will Data-Driven Feedback Loops Replace Technology as the Moat 36:08 Who Wins the Data Labelling Market? Who Loses? 38:23 Is Revenue BS in Data Labelling? Are Players Calling GMV Revenue? 45:20 Why is SaaS Dead in a World of AI? 51:23 Will the Phone be the Primary User Interface to an AI World? 57:07 Quickfire Round
Transcript 12,409 words · 2 speakers
Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.
0:00 Speaker 00 I think the era of data labeling companies is over and it's now the era of research accelerators. All knowledge work is going to be automated. It's only a matter of time. I don't see an AI bubble. These models are incredibly powerful today. SAS as we know it, I think is over. I think it's completely over.
0:18 Harry Stebbings? This is 20 VC with me, Harry Stebbings, and I'm so excited for the show today. So I'm fascinated by the data labeling market. McCord, Surge, Turing, Invisible. There are seven players that I know of that do over a hundred million in annual recurring revenue. All of them have more than fifty percent of their revenue from two customers. The way they use the term revenue is also sometimes super unclear versus GMV. There are bluntly just a lot of questions. Today, I do not pull any punches with the man leading one of these companies, Jonathan Sedat, founder and CEO of Turing, a company that he has scaled to over three hundred and fifty million in annual recurring revenue,
0:59 Harry Stebbings? raising two hundred and twenty-five million dollars in the process and now being a profitable company. This is an amazing discussion and as I said, the tough questions get answered. But before we dive into the show today, are you drowning in AI tools? ChatGPT for writing, Notion for docs, Gmail for email, Slack for comms, and you're constantly copy pasting between them all, losing context and losing time. This is the AI productivity tax and it's killing your output. At 20 VC, we're all about speed of execution and Superhuman is the AI productivity suite that gives you superpowers everywhere you work. With the intelligence of Grammarly, Mail and Coder built in, you can get things done faster and collaborate seamlessly.
1:45 Harry Stebbings? Finally, AI that works where you work, however you work. Superhuman gets you from day one with zero learning curve and it's personalized to sound like you at your best, not like everyone else using generic AI. Get AI that works where you work. Unlock your superhuman potential. Learn more at superhuman.com/podcast. That's superhuman.com/podcast.
2:10 Harry Stebbings? And just like a superhuman gives you superpowers in your inbox, Vanta gives your company superpowers in security and compliance. Customer trust can make or break your business, and the more your business grows, the more complex your security and compliance tools get. It can turn into chaos, and chaos isn't a security strategy. That's where Vanta comes in. Think of Vanta as your always-on AI-powered security expert who scales with you. Vanta automates compliance, continuously monitors your controls, and gives you a single source of truth for compliance and risk. So whether you're a fast-growing startup like Cursor or an enterprise like Snowflake, Vanta fits easily into your existing workflows so you can keep growing a company your customers can trust.
2:54 Harry Stebbings? My listeners can get $1,000 off Vanta by going to vanta.com/20vc. That's V-A-N-T-A dot com forward slash 20vc, 20vc for $1,000 off Vanta. And like Vanta builds trust with automated security, AngelList drives momentum by helping you find and fund breakout startups. If you're listening to 20vc, you know we have a really frickin' high bar. Well, AngelList is the modern platform used by the best-in-class venture funds, where over 40% of top endowments and banks are LPs. Their customers include a top-five venture firm, 20vc, and they now have, check this out, $171 billion of assets on the platform.
3:38 Harry Stebbings? They combine an all-in-one software platform with a dedicated service team that moves as fast as you do. One manager said this awesome quote, "AngelList feels like an extension of my fund." Another said, "AngelList gives me total peace of mind, the attention to detail, lightning-fast response time, and just real sense of ownership from the team are exactly what I need to stop worrying about back-office ops." So if you're starting a new fund, don't be a moron, just use AngelList. They're incredible. Head over to angellist.com/20vc to learn more. You have now arrived at your destination. Jonathan, I've been so looking forward to this. Thank you so much for joining me in person. It's such a treat to do it in person while you're in London. Thank you for having me, Harry. Now, I want to start with a little bit of definitions
4:21 Harry Stebbings? because everyone thinks they're talent marketplaces, and then everyone pushes back on talent marketplaces. How do you describe it? And why are we not dealing with talent marketplaces anymore?
4:33 Speaker 00 I think of a talent marketplace as something that's basically matching talent to something. Maybe it's an opportunity. So Turing is not a talent marketplace. What we do at Turing is we're training superintelligence. We work with seven out of the eight frontier labs. To get to superintelligence, you need research, compute and data. Research, the labs do in-house with OpenAI, Anthropic, DeepMind, et cetera. For compute, we have Jensen to thank and maybe Nvidia as well. On the data side, Turing powers the data pillar. On the data side, there's been a significant shift in the last couple of years. So a few years back, the models weren't quite smart enough. And as the models have gotten increasingly smarter, the data needed to improve them has become harder to generate.
5:18 Harry Stebbings? And this is because it's more sophisticated data that's required to improve the models. It's like vertically specific people in task and workflows that isn't so obvious, like cat pictures.
5:28 Speaker 00 That's correct. And there's a shift in the data going from simple to complex. I mean, let's take coding, for example. A few years ago, the kind of data set a contractor could generate might look like, hey, write a Python program to sort some numbers. Today, the data that's generated might be, write a B to B marketplace app that connects doctors with patients. And write it on for Android with Kotlin Java, write it for iOS with Swift and write it on the web, like with NextJS or something. That's the complexity. So there's a shift in going from simple to complex. So it's no longer the kind of data that low skilled, medium skilled contractors can generate. You need expert humans in every domain.
6:08 Speaker 00 The second shift is we've gone from teaching AI to take tests and pass tests to teaching AI to do real work. It's less about having AI pass the bar. It's more about can AI do the job of a lawyer? Can it do the job of a privacy lawyer, a compliance lawyer, a paralegal? Having AI be good at doing economically valuable work, that's a shift. The third shift is we've gone from chatbots to agents, right? Like we started off with ChatGPT where you're asking questions, getting answers, which is great. But now it's about the models becoming agentic, where they can execute complex, multi-step workflows in a real world business setting. And the type of data you need for that is totally different.
6:52 Harry Stebbings? How is that different? That's so interesting. So in the transition from chatbots to agents, how does the data required change with that transition?
6:59 Speaker 00 When you are training a chatbot, you'd usually do a lot of SFT and RLHF. SFT, where you're giving the model input prompts and output completions, you kind of teach the model to imitate experts. With RLHF, you're basically teaching the model to produce responses that a human would tend to prefer. RLHF is used to train what's called a reward model, and then the model is trying to produce completions that give it a high reward. With agents, and let's define an agent. I mean, different people define agents in different ways. I would define an agent as something that's capable of taking action in the real world or in the physical world, something that's executing a multi-step workflow, calling different functions.
7:43 Speaker 00 The agent could be operating a computer or making backend API calls to actually do stuff. You might have an agent to file your taxes. You might have an agent to prepare your monthly financials. So to train an agent, you would also want to teach the model how to do tool use. So you teach the model how to call other functions, how to use other applications to be more leveraged. Today, the dominant paradigm is reinforcement learning. And oftentimes, these agents are trained through reinforcement learning, where you'd build what's called an RL environment, which is like a mini world model for business. Yep. And in that RL environment, you have input prompts, output verifiers, and you'd have the full system state tracked
8:28 Speaker 00 along with the data model. Let me give you an example. Imagine a workflow for a salesperson that an SDR would go through, where before a sales call, let's say the salesperson has to research the prospect, look up Salesforce to see whether somebody from the team has already spoken with this human, and maybe if needed, looking up this person's contact information, maybe using ZoomInfo or something like that to reach out to them. This required this human to use three different tools: LinkedIn, Salesforce, and ZoomInfo. In an RL environment setup, you'd create a mini world model with clones of these applications that are created with a fake database with like synthetic data, and the prompt might be...
9:09 Speaker 00 Prepare for a call with this person and then after the call is done, update Salesforce. Right? Like, let's say that's the prompt. And you have what's called a verifier to check whether the agent completed the task. And in this case, and this is where I think AI is kind of beautiful and somewhat magical. You set up the agents in this environment and the agent is going to try different trajectories, try different tool calls to try to complete the task. And you would set this up so that the curriculum is optimally defined. The curriculum is the set of tasks that you have this agent do. If it's too easy and the agent completes everything, the model doesn't learn much. If it's too difficult, the model doesn't learn much. You'd ideally want like the the right mix where the model is kind of getting positive and negative feedback.
9:53 Speaker 00 It's very similar to the technique that AlphaZero used in Mastering Go, where the model played against itself. It's kind of like a form of synthetic data because the agent is trying different approaches by itself. But it's humans. I mean, in this case, at Turing, we create these RL environments at massive scale for every workflow you can think of across every function, across every industry.
10:15 Harry Stebbings? And so you create the RL environments that create the data that then allow the models to train to have further use cases like that?
10:22 Speaker 00 Correct. And we create RL environments for every industry you can think of, retail, healthcare, life sciences. Imagine like this four dimensional matrix where the first dimension is every industry, like financial services, retail, healthcare, podcasting, maybe it's one of the dimensions. The second dimension could be every function, software engineering, marketing, sales, finance, et cetera. The third dimension could be every role in that org shot. Let's say in sales, there was SDR as like a role. The fourth dimension is a workflow that a human goes through in that role. You can think of every role a human has as like a composite of workflows. And we are creating RL environments for every workflow, for every role, in every function, in every industry. That's like thirty trillion dollars of knowledge work.
11:03 Harry Stebbings? Is that possible to have that breadth and quality?
11:06 Speaker 00 Yes.
11:07 Harry Stebbings? How?
11:08 Speaker 00 Time and lots of money.
11:11 Harry Stebbings? Because I was speaking to candidly one of your competitors' board members the other the other day in prep for this. And he said, the big thing that we all got wrong was we are so in innings one of the acquisition of verticalized data. There is so much room to run in the data acquisition for dental, for SDRs, for product managers, you name whatever function you want. Do you see us very much in innings one of the data acquisition for these very specific vertically focused workflows?
11:39 Speaker 00 Absolutely. It's innings one, and I believe in slow takeoff. I'm sorry to pour cold water on all the AI doomers that might be listening to this, but we are not in a rapid takeoff scenario. I believe in slow, steady takeoff for AGI and eventually superintelligence. So we're still in innings one. It's going to take a while before we get all of this data into the models.
12:03 Harry Stebbings? When we think about then, the breadth that we go after and us specializing in RL environments, just so I understand, like the marketplace that we sit in, because there is your McCausers, your Serge's, how do you differ for those, for people who are wondering, hang on a minute, I thought they were all one?
12:18 Speaker 00 Turing is a fundamentally different animal. So what we do is we're training superintelligence for all these frontier labs. To get to superintelligence requires research, compute and data. The data needs have significantly changed. It's more complex data rather than simple data. It's more real world data, data that touches how real humans do knowledge work. And you need data to train these agentic systems. What the labs need in a partner in this new world is somebody that has research DNA that could be a proactive research partner for them. Because these paradigms keep changing. Last year at this time, we were not talking about reinforcement learning at all. But then two things happened.
13:01 Speaker 00 O1 dropped in December, DeepSeek launched in Jan, and now it's all about RL environments. It's not just imitation learning, it's also reinforcement learning. So the labs need a data partner that's more research oriented. Second, the labs need a data partner that also touches the real world. So at Turing, We don't just generate data to train the models for the Frontier Labs. We also work with enterprises. We work with Disney, Pepsi, BlackRock, Fiserv, Johnson and Johnson, to build fine-tuned custom models to solve real-world enterprise problems for those enterprises.
13:37 Harry Stebbings? So this is like FDEs that you send in to go and build custom models?
13:40 Speaker 00 Correct. So we touch reality. We know where the models break in the real world.
13:45 Harry Stebbings? How much of business is that, the FDEs deploying custom models versus more horizontal?
13:49 Speaker 00 So the horizontal business is bigger, but this business is also fast-growing. And third, you need a platform with the world's smartest humans on it, as well as experts in different domains, so that you can build these RL environments. And you need that platform to be really good at sourcing talent, vetting talent, matching talent, and generating data. I think the era of data labeling companies is over. Turing is a research accelerator, and it's now the era of research accelerators. The labs want to work with a proactive partner that can think about what types of data are likely to be helpful for these models and can make recommendations to them.
14:29 Harry Stebbings? Why would you need a custom model? When you look at a lot of the customers that you mentioned there, for the ones where you have FDEs who go in and build custom models, what is the reasoning around that? And is that a temporary moment in time? Or is that a permanent requirement from them for a certain reason?
14:43 Speaker 00 I think it's a permanent requirement. I'll give you an example. Let's pick an insurance company. For insurance companies, two really important problems they have to solve are underwriting and claims processing. Let's pick underwriting, for example. So with underwriting, the problem statement is you might get multiple types of unstructured medical data. It could be somebody taking a picture of their medical history on their smartphone, or it could be some OCR data from somebody's medical history, or data in PDFs, et cetera. And a human has to look at that person's medical information and then decide, is this person high-risk, medium-risk, or low-risk? What medical conditions do they have? Do they have cardiovascular? Do they have renal? And how do you price insurance for somebody like this?
15:28 Speaker 00 Do you even take them on as a client, if you're an insurance company? Now, this is a problem that an L L M can solve really well with a human in the loop system. You may not need a trillion parameter world model to do a task like this. In fact, there's lots of research that shows a smaller language model will actually be faster and more accurate at a task like this than a giant world model. And the insurance company also may not want their data to go back to like a frontier model. So oftentimes in these cases, what we would do is we would work with a frontier lab, take one of their smaller models, maybe maybe something in the half a billion parameters
16:09 Speaker 00 to 10 billion parameter regime, have an A I system built that's on prem with the customer that's fine tuned on that enterprise's proprietary data. Because this insurance company might have data from like the last decade of humans making judgments, right? You want to make use of that data, but you don't want to help other competing insurance companies with your own data. So normally for these cases, you would have a smaller fine tuned model that's fine tuned on your proprietary data, distilling your proprietary human knowledge into the models. That human underwriter that's been doing this job like has a lot of institutional knowledge in their brains. You'd want to distill that into the L L Ms.
16:51 Speaker 00 And this human that I mentioned, who's doing this job of underwriting, they might be using other internal tools inside this insurance company. You might want to automate those tool calls in the agent. So you'd build basically a version, almost like a chat G P T agent that's smaller and fine tuned for that specific workflow and use case. I do think this will become more popular across the board and it'll be a big market for the frontier labs. Like if you want like a general purpose assistant, I think you need a trillion parameter model, right? Like that can answer anything to be a universal assistant.
17:22 Harry Stebbings? Is that a good business for you? When you think about taking someone else's model, retrofitting it to a business, doing a lot of custom work with your own engineering teams in those businesses, is that a good business?
17:33 Speaker 00 It would be a good business. Like we're still early, but it's growing pretty fast. We have this unique vantage point where because we are
17:42 Speaker 00 Generating data for all the frontier labs, we get to see a glimpse of the future before it arrives. And the glimpse of the future that I see is that all knowledge work is going to be automated. If a human's job involves looking at a computer, analyzing what's on the screen, using different tools, using a keyboard and a mouse, it's going to be automated. It's only a matter of time. I mean, these computer use agents are going to keep improving over the next decade. That's thirty trillion dollars of digital knowledge work.
18:13 Harry Stebbings? My question to you is, I, you know, I spend a lot of time with very large companies, mostly where I speak to them. The thing that I'm astounded by is we hugely underestimate the pace of AI progression in terms of technological capabilities. But consistently, what I see is the laughable state of internal data, internal processes. I mean, Jonathan, these guys are so far off adopting Slack and Notion, let alone building custom models and embracing the latest AI tools. I respectfully push back on the all knowledge work automated, maybe in twenty years, but not in a ten-year timeframe. Am I wrong?
18:50 Speaker 00 What do you think is the biggest constraint or obstacle?
18:53 Harry Stebbings? The inability for them to try and implement new tooling.
18:58 Speaker 00 But what if the cost is too high? If they didn't do that, if the hypothetical insurance company that I told you about, if there was a competitor of theirs that could operate with one hundredth the headcount while delivering a better experience to their customers by pricing insurance deals better, making more money from insurance premiums, less claims payouts, they'll get their lunch eaten.
19:21 Harry Stebbings? I think they will. I think you'll see this transfer of value from old incumbent who can't adopt new tools to startup company, hence why we invest, who is eating their lunch absolutely.
19:32 Speaker 00 Ah, I see. So your theory is that the incumbents won't adapt and we'll just be a forest fire and you'll just...
19:39 Harry Stebbings? A hundred percent. We will be on a ten to twenty year decline of incumbents who are unable and unwilling to adopt new tools because of data, because of permissioning, because of internal buying processes. I mean, I go to a European bank, you will be astounded by how bad it is internally, the poor quality of the processes of buying technology. It's abhorrent.
19:59 Speaker 00 I have a hypothesis. My hypothesis is that companies will be very slow with back office automation, but in the front office, for example, I speak with financial services clients in New York, like some of the biggest, biggest companies. I speak with the C suite of these companies and they are extremely interested in applying AI to help them make better investment decisions because it directly translates into helping them make more money. It's a lot easier to like convince people to use a piece of technology to make more money than to save money. And in financial services, it's pretty brutal, right? Like it's kind of an efficient market. If there is alpha to be found in like how you can allocate capital better, make investment decisions better, figure out what opportunities
20:44 Speaker 00 to invest in, price deals better, you'll get killed if you're not at the bleeding edge. And I've heard Mark Chen, the head of research at OpenAI, say this about how financial services is usually at the bleeding edge among all the other industries in the S and P 500. But even they are like about two years behind usually relative to the state of the art. So I agree with you that in back office automation, it'll probably be very slow and it'll probably be the upstarts that'll do things well. I think the change management will be too slow. But I'm optimistic about front office, especially in financial services, life sciences, pharma, where if you can like accelerate the time to discover
21:25 Speaker 00 a drug or to get to a molecule, if you can help somebody win in like the main thing that they care about in their industry, I think they'll adopt it faster.
21:35 Harry Stebbings? I'm always told by a dear friend Rory O'Driscoll at Scale, I don't know if you know Rory, but fantastic investor. And he always says to me, listen, value generation from AI is fundamentally dependent on one simple question. Will we see the transfer of budget from human labor to AI technology? And if we see that transfer of budget, oh my God, the $30 trillion that you said, it does. And if we don't, we operate in maybe a slightly larger software technology budget world, but by no means a world that we can have the valuations and the money that we have going in. When you look at that, are there any areas truly today where you're like, we have seen the full transition from human labor budgets to AI technology budgets?
22:18 Speaker 00 I think the transfer is pretty high in areas like customer support, copyrighting, SEO, like some of these marketing related areas. As you would expect, the transfer is faster in these low risk to fail areas, like when it's relatively easy. But I'd encourage your listeners to look up GDPVAL, which is this paper by OpenAI, where they measured the impact of today's AI models in automating all types of economically valuable work. It's a lovely piece of research. I'd encourage everybody to read it. Where they did this study where they looked at, I think, like nine verticals and forty
23:01 Speaker 00 four occupations. They took like a very diverse sampling of different types of knowledge work, everything from financial services to real estate to healthcare to law. And they took very specific occupations. And in those specific occupations, they took real tasks where like a real deliverable has to be produced. Imagine an engineer, like a civil engineer, creating a blueprint for a building they're about to build, or somebody who's at the set of a movie studio, coming up with a schedule for how you organize your crews, like real work, right? And for coding, you can imagine like a real world software engineering project. And they saw that today's models were quite good at achieving parity with the best human experts in that field. We were roughly at like, I say we, I don't
23:46 Speaker 00 know which side am I on? I'm on the side of the AIs or the humans?
23:48 Harry Stebbings? I think you're on the side of the AIs.
23:50 Speaker 00 Oh, really?
23:50 Harry Stebbings? Yeah, yeah.
23:51 Speaker 00 I mean.
23:51 Harry Stebbings? From that positioning, that would infer so.
23:53 Speaker 00 Yes, yes. But Turing is like a blurry line, right? Like as the CEO of Turing, like passing the Turing test is about not being able to tell the difference. So what I noticed was about fifty percent of the time in GDPVAL, like the best models were producing work that was indistinguishable from a human expert, which is remarkable. And kudos to OpenAI, where they also flagged that the number one model was Claude four Opus, although GPT-5 was quite good also. And this was for relatively simple tasks, like a task requiring one single step, whereas in the real world, if I give Give you a task to do a certain project, you won't just go off and do it. You might ask for clarifying information. You might do other things to acquire more context.
24:36 Speaker 00 You might go brainstorm with other other humans to do that task and you would do it in a sequence of steps. So there's more room to go, but I think we are well on our way to AI eventually automating all types of knowledge work.
24:50 Harry Stebbings? What happens in that world? If AI automates all types of knowledge work, what happens then?
24:55 Speaker 00 Three things that will happen. First, we'll all have the potential to be a hundred X more productive. Today, I'm able to run one company. Elon can run maybe like five companies. But in a world where I'm a hundred X more productive, maybe I'm able to run a hundred companies. Elon maybe runs six hundred companies. Every human will just be so much more leveraged. The nature of a job itself could change. Today, we are accustomed to the idea of like one person doing one job. But people could be doing multiple jobs at the same time. People could be running different companies at the same time. The second implication, I think, is it's going to be wonderful for entrepreneurship.
25:34 Speaker 00 So you, Harry, I think are going to be very happy because today for a lot of ideas, founders are intelligence constrained. I think of being capital constrained as a form of being intelligence constrained. For example, if you pick a therapist who wants to start a mental health startup. Today, that founder would have to raise at least a few hundred K, if not a few million, to recruit some software engineers. Maybe a marketing person or a growth person, maybe a product manager. But in a future where AGI exists, this person will recruit a marketing GPT, a software engineer GPT, a PM GPT. And get off the ground for a lot less capital. A million flowers will bloom, lots and lots of non-technical founders will start companies.
26:15 Speaker 00 We'll see a broader distribution of founders than just those who live in London or Palo Alto. Who are kind of connected to these pools of capital who might start companies, which I think is wonderful for the world.
26:26 Harry Stebbings? Do you think we will? What I mean by that is there's six and a half million people today in the UK. That's a working population who actively do not work because of an inability to work. I'm not going to get into the analysis. around that because I'll get in trouble for it. I think we grossly overestimate the intelligence of the general population. And I know that sounds incredibly arrogant, but most people say, actually, a lot of people just don't want to work and are not at the level of recruiting GPT's assistance. Do you not worry that it will widen the chasm between those that have and those that haven't?
27:00 Speaker 00 I'm an optimist and I think the opposite will happen because what we're really doing when we're training superintelligence is you're basically training intelligence as an API. And what's the alternative to that? It's like hiring a human to provide you with that intelligence. That human is quite expensive. And if anything, that creates an even broader gap between the haves and the have nots. Whereas for twenty dollars a month, if you had access to the smartest experts in coding, in STEM, in sales, in marketing, I feel like more people will be able to start companies and produce active valuable work. I believe that when we have access to superintelligence, I firmly believe this. We are not all going to chill out on a beach somewhere and contemplate, what do we do next?
27:45 Speaker 00 I mean, we humans, like we are tool builders, we are problem solvers. We'll solve problems at higher and higher levels of abstraction. In a world where we have AGI, we'll just solve much more exciting problems. Maybe we'll cure diseases, reverse aging, maybe go to the stars. I don't think we'll be bored.
28:02 Harry Stebbings? I'm glad. I often hear the UBI and we're going to sit and write poetry and I'm like, I think that might be a little bit challenging. When technology is not the moat, what is the moat? I had the founder of Base44 on and he said ninety nine percent of code in that year will be written by AI. Technology is no longer the moat. What is the moat in that world?
28:21 Speaker 00 I think one moat will be data-driven feedback loops. For example, one reason Google had such a great lead in search for a while was these data-driven feedback loops that come from people using your product and generating data that gives you, the algorithm developer, a high-quality gradient for which direction to step in. So the importance of PageRank was known. The recipe for ranking. Making search results was well known among Google, Yahoo, Microsoft and a few others. Obviously, people move around these companies all the time. But the advantage Google had was because everybody preferred Google and liked that search engine. You saw a much more representative set of queries.
29:06 Speaker 00 You had data from click stream, from the click stream of what results people were clicking on. That helps your algorithms improve at a much faster rate. I think data-driven feedback loops will be key for all types of enterprise applications also. Today, OpenAI and ChatGPT has a good data-driven feedback loop. In enterprises, again, I think it's wide open. Whoever is deploying the right custom, fine-tuned models and agents for specific workflows or roles or functions or companies. If you get in first and solve a customer's problem really well, you start getting that flywheel going where you will discover first where the models don't work well. And you will use that data to work with a company like Turing
29:51 Speaker 00 to generate additional data to plug that gap, so you will improve. This is what I mean by it's important for the models to touch reality. I feel like the models have touched reality in consumer. We haven't yet touched reality in enterprise. And the only way we'll improve is by deployment.
30:06 Harry Stebbings? And that deployment is fundamentally predicated on handholding, correct?
30:10 Speaker 00 Yes, handholding. And I feel like there's still a lot of first-mile schlep and last-mile schlep that needs to be handled.
30:17 Harry Stebbings? What does that mean, first-mile schlep and last-mile schlep?
30:20 Speaker 00 When I say first-mile schlep, I mean for that underwriting co-pilot example that I gave you for that insurance company. I painted a pretty rosy picture of like how you take this model, you fine-tune it on your proprietary underwriting data. In the real world, it doesn't work that way.
30:36 Speaker 00 First-mile schlep is, let's say I'm talking to the CEO of this insurance company or the CTO of this insurance company. They'll say, our data is a mess, it's in silos, it's super fragmented. Some of the data is in spreadsheets, some of the data is in a file that Bob has. And Bob doesn't work here anymore, right? The data kind of is all over the place. You first have to acquire the data, convert the unstructured data into structured. Data into a format to fine tune L L Ms. You might wanna set up good infrastructure for evals. You'd wanna create good evals for the models or agents. You might wanna build a workflow designed for partial autonomy. So this human underwriter that's about to use this model to evaluate these medical histories.
31:18 Speaker 00 You might wanna build a cursor like interface for them so that they can work alongside the A I to do their job. Also, training the humans in these new workflows. You wanna make sure you're collecting data the right way. The way, for example, we do deployments is like a tandem system where you'd have a human and A I doing the same job for a period of time, where a manager can see the output of both. If the agent is right and the human is wrong, you train the human. If the human is right and the agent is wrong, you've created a data point to fine tune the next iteration of the agent. So the agent is steadily improving over time.
31:54 Harry Stebbings? If the agent is right and the human is wrong, why don't you just fire the human?
31:58 Speaker 00 I mean, it depends on at like what frequency, right? Like you you track things like precision and recall, like you'd wanna analyze this over a period of time. Like you wouldn't fire them over like a single mistake.
32:08 Harry Stebbings? Bit harsh.
32:09 Speaker 00 Yes.
32:10 Harry Stebbings? What's the margin on that business?
32:12 Speaker 00 It varies like we're also in the early innings of figuring out how to price that. Today, like we do it in a relatively simple way where we're just billing these things for time. I don't think that's the right way to do it. Like we'll switch to a more value oriented pricing model at some point. Right now, we're just laser focused on the frontier eight labs. So enterprises for us is is a longer term play.
32:37 Harry Stebbings? When you look at revenue numbers in this space, a lot of people shout back, they're not revenue numbers, they're G M V. Given our understanding now that, Bunny, there's no talent acquisition from your business. It's all an R L environment creation business. When you look at the other announcements of alternative providers, can you help me understand? Are they revenue or are they G M V? And is there mislabeling being done here?
33:02 Speaker 00 So I don't want to comment on other companies, but we think about revenues differently. Like the way we think about revenues is more in terms of GAAP revenues.
33:13 Harry Stebbings? It's like traditional revenue numbers. I'm an investor, Jonathan. Essentially, I'm trying to understand and learn from you of how I should weight revenue in today's AI world versus a previous historical world. Like, should I be impressed by these revenue numbers or should I not?
33:27 Speaker 00 I think it depends on the type of revenue. Obviously, these are not SaaS ARR numbers, right? These are not those types of revenues. This is a different beast. I think this requires thinking from first principles. The revenue here is reoccurring in the sense that oftentimes when you're working with a lab on helping the models improve in some area, and I'll speak to Turing, I don't want to speak to other companies. When we are helping a lab, let's say, improve their models for coding or multimodality or tool use or are working on RL environments for automating all types of professional knowledge work, it's usually a reoccurring project. Projects will start, projects will end, and as long as you're doing a good job,
34:10 Speaker 00 there is lots and lots of demand, but you have to consistently keep doing a good job. It's also important to be a trustworthy partner to the labs. Like, we take secrecy very seriously. We make sure that our projects are all firewalled between labs. Oftentimes, even with teams within the labs, like sometimes like that's the level of secrecy that you'd need. I'm reminded a little bit about how I've been told Foxconn operates. I mean, I don't know anything about that, but I've been told that they have different floors where, you know, maybe they're on one floor, the iPhone is getting made, and on another floor, maybe a Pixel phone is getting made. And obviously, you have to like firewall all of that.
34:46 Harry Stebbings? Of the eight largest providers, do they not spend with all of you?
34:50 Speaker 00 They spend with a handful of companies. They do that to have some level of resilience. And I imagine there's some price benefits to having more than one person that they could work with. But I think the resilience piece is important. I mean, we know what happened with the, you know, when the scale investment happened, the labs did benefit from having other partners that they could work with. I would say it's a small handful. It's a small handful that are trusted. And of course, there's probably a giant pool of smaller startups, but it's a small handful of big companies in the space.
35:24 Harry Stebbings? Which one do you worry about most?
35:26 Speaker 00 So this is just a big, big market that's growing super fast. I'm excited for like all the companies in the space. I feel like different companies come into this world with a different DNA.
35:39 Harry Stebbings? Which leader do you most respect?
35:41 Speaker 00 Sam Altman, Elon Musk.
35:44 Harry Stebbings? Of the data providers, Jonathan?
35:46 Speaker 00 Of the data providers?
35:46 Harry Stebbings? I'm pushing you to. This one, I'm gonna get a name.
35:51 Speaker 00 I mean, I have a lot of respect for Alex Wang from Scale AI. I feel like Alex and Scale were prescient in seeing the importance of data. And I admired how, having started in autonomous labeling, like how they kind of navigated the ups and downs. Yeah, I really like the way he operates as well. I feel like there are certain elements of leadership that I think I share with him. And I think he did, he did a great job for, for Scale.
36:20 Harry Stebbings? How did Scale being acquired impact Turing's business?
36:24 Speaker 00 We just got flooded with a lot of demand. And we've also amped up pretty significantly in multimodality. Multimodality was something I think Scale was quite strong in. Multimodality is teaching the models to operate well with not just text, but audio, video, image, et cetera. Outside in, I've heard that because of their roots in autonomous labeling, like they were quite good in multimodal stuff. It was, it was good primarily from just increasing demand. I feel like they were the company that had been working in the space the longest.
37:00 Harry Stebbings? Do they have a business left? Again, I mentioned Roy O'Driscoll. I think he said in a show with me recently that there's this kind of carcass or husk left behind. If everyone benefited from their being bought, they can't be doing that well.
37:13 Speaker 00 I don't know enough about their business, like in terms of.
37:17 Harry Stebbings? Do you pay attention to competitors?
37:19 Speaker 00 I pay attention to competitors in terms of things that they do well and when there are any significant learning opportunities from them that could help us serve our customers better.
37:31 Harry Stebbings? Do you worry about revenue concentration? You said about kind of eight of the biggest labs. Say if you look at like an OpenAI, they have, I don't know, whatever it is, a hundred million. You all know these numbers much better than me, but say a hundred million paying customers. I'm just taking a 10% on a billion people, but give or take a hundred million, whatever. And then you look at, say, a business like ours here, where there's like seven core customers. How do you feel about revenue concentration?
37:54 Speaker 00 The last time I checked, I was told that for Nvidia, 39% of their revenue comes from two clients. And roughly 50% was like four clients. So in this market, I don't worry that much about revenue concentration. I think we are in the same boat as Nvidia in that there will be lots and lots of spend from these big eight companies. Look at the scale of the spend, right? Like Stargate is like a hundred billion a year investment on compute. There's going to be significant amounts of spend on compute, energy and data. It's a little weird to have this level of concentration, but things could change. I think it's also possible that governments spend even more. I think it would make sense for governments to build their own internal versions of some
38:39 Speaker 00 of these models, which would require proprietary data again to be collected.
38:43 Harry Stebbings? Do you not think they'll have to? And what I mean by that is like we'll see sovereignty of models. Like I do not think there's any way you'll have the German healthcare system working with American model providers.
38:53 Speaker 00 I think you're right. I think it'll, it'll be necessary.
38:55 Harry Stebbings? In that world, do you provide that FDE service to governments?
39:00 Speaker 00 Yes, the work that we're doing in not just training superintelligence, but in deploying superintelligence is with that goal in mind. For these governments, I imagine we'll help them not just with fine tuning their models with data and evals, but also helping them with the first mile schlep and the last mile schlep to make these systems actually work. And you might want to have like full control over what type of human data is going into these models. Like if you're the German government, presumably you want German nationals to be the ones that are contributing data, whether it's for SFT or reinforcement learning.
39:32 Harry Stebbings? So I'm worried that we are potentially not going to see AI deliver the immediate revenues that we've promised, and we're going to go through a kind of cooling period, which everyone kind of suggests and thinks that we're going to go through in the next six to eighteen months, which is, as I said, it doesn't hit the revenues that we said it would, and kind of the AI bubble deflates slowly. To what extent do you think that's possible, or we'll see this continuing gradual increase, as we kind of touched on there?
40:00 Speaker 00 I don't see an AI bubble. These models are incredibly powerful today. GPT-5 is like fucking awesome. I don't know what people were talking about when they're talking about, you know, I know there was some chatter. I think we've just gotten used to magic. I feel like, A, these models are incredibly powerful today, and they're the worst they'll ever be. They're only going to keep improving. I say that about the Gemini Pro models, the Grok models, the Claude models. Like, these models are amazing.
40:26 Harry Stebbings? Yeah.
40:26 Speaker 00 And there is a very significant model capability overhang. By that, what I mean is, the models are capable of X, but what we are getting out of the models is X minus delta. With the right agent X scaffold around these models, in terms of the right system prompts, the right user prompts, giving the models access to the right context, teaching the models how to acquire additional context, teaching the models how to use the right internal tools, there is significant amount of capability that can be unlocked with today's models. For example, you, Harry, I imagine, when you do an interview with somebody, one of the things you probably do is you apply your secret sauce to pull out the right clips from the interviews, to like, what to highlight, what are the catchphrases, what will drive
41:10 Speaker 00 more engagement. That can be done by a model with the right agent X scaffold that's fine-tuned on all the work that you've done in the past.
41:18 Harry Stebbings? I wish. Every weekend, I go through every single show.
41:22 Speaker 00 Yes.
41:22 Harry Stebbings? And I pick out fifteen to twenty clips per show, and then I make notes on each one.
41:26 Speaker 00 Yeah. Are you saying, Harry, that you want to use Turing?
41:28 Harry Stebbings? If you could fucking make it work, dude, I'll pay you a lot of money.
41:31 Speaker 00 When we, yeah, maybe we should, we should partner.
41:33 Harry Stebbings? That'd be great. Seriously, every weekend I spend probably three hours per show, three, definitely 12 hours a weekend doing that.
41:40 Speaker 00 I think there is this model capability overhang where the full potential of the model has not been unlocked by humans yet. And no, I don't think there's an AI bubble. I think there are some growing pains.
41:52 Harry Stebbings? What are the growing pains?
41:54 Speaker 00 I think everybody keeps citing that MIT report about how 95% of pilots fail. But because we are in the business of deploying AGI in enterprises, I can tell you like why I think that happens, which is one of the growing pains. Step one is that most enterprises need to do some work to structure their data in the right way. Again, that first mile schlep has to be done. Second, you should surround the model with the right agentic scaffold that I just described. The right prompting, the right context engineering, the right internal tool calls that you should teach the model to call. All of those have to be distilled into the models. You need really good evals. You also need a workflow designed for partial autonomy.
42:36 Speaker 00 Andre Karpathy articulated this first when talking about why Cursor works so well. Because it's not designed for full autonomy, it's designed today for partial autonomy, for humans to collaborate with the AI to do that specific task. So that Cursor for X needs to be built for every role, for every workflow to like help humans work more easily with the models.
42:57 Harry Stebbings? Does every role need to go through that pathway of Cursor for X before it goes to full autonomy? Or are there some roles like customer service where it just goes to full autonomy?
43:06 Speaker 00 Some roles where you can see that the models are quite good at matching humans. I think we don't need that intermediate step. There are certain roles where by virtue of how the models are trained, where they're pre-trained with tokens on the internet. And then of course with talent from research accelerators like Turing, that's fine tuning the models. But there are certain types of roles where the tokens from the internet give it sufficient intelligence to do the job well. Customer support is an example. But if you pick other roles, like for example, if you picked the role of, let's say, an AI researcher, or you picked the role of a lawyer. Specializing in venture capital financing. It's possible there's not enough of those tokens on the internet. So the models will be relatively weak there out of the box.
43:51 Speaker 00 And also, the way Wilson Cincini does financing might look different from the way Akuli does it. Maybe they have their own way. So you might want to fine tune them on your own proprietary data. Distill the proprietary intelligence of humans working there. So for those things, like you may need to do some fine tuning. The models may not work very well out of the box.
44:11 Harry Stebbings? A lot of people suggest the circular deals between some of the large providers suggest the strains in the ecosystem or the bubble like tendencies. Do you think that's fair or not?
44:20 Speaker 00 I've sort of categorized the world into two classes. Class one is those that believe in AGI. We, I mean, let's call it the AGI-pilled group that believe that we are on the path to getting to AGI. And let's define AGI as an AI system capable of at least matching humans in almost all types of intellectual knowledge work. Then there is a world, maybe another category of people that don't believe this will happen. We'll hit a wall, right? And in the past, there have been other AI paradigms where we did hit a wall. For the camp that believes in AGI, and I believe in AGI, unsurprisingly. But I love AI and it's been my passion for like the last 20 years. I really believe that we will get there.
45:01 Speaker 00 If you believe that, the grand prize is so amazing, right? Like if you've solved intelligence, you've solved all of humanity's grandest problems. From curing diseases to potentially pausing aging, to like interstellar travel, to energy. All of our problems are intelligence constrained. So the prize is so large. I mean, whoever wins the superintelligence race will probably win search. Will probably win consumer devices. Will probably win operating systems. Will probably win cloud, like business productivity software. It's like the prize is so massive that it's worth placing big forward bets in these areas. Because the cost of not winning is too high. Whoever wins AGI would also probably win social networking. So you can see why the big eight are excited about it.
45:45 Speaker 00 Because it's, you're playing for everything. It's like whoever wins this could be responsible for that $30 trillion of knowledge work.
45:51 Harry Stebbings? Well, if you're Zuck, you spend a hundred billion dollars on it. If you lose or fail, likely everyone else will fail, in which case you're behind just like they are. And you've lost a hundred billion dollars, which isn't a huge amount of your free cash flow. Maybe twelve to eighteen months of free cash flow. If you don't spend that hundred billion dollars and someone else does and wins, you lose two trillion, three trillion dollars of market cap.
46:12 Speaker 00 Correct.
46:13 Harry Stebbings? You have to play.
46:13 Speaker 00 You have to play. Imagine if somebody built a more engaging social network. And social networks have just one unit, which is attention. I mean, we all only have like maybe, maybe four to five hours a day to spend on an app. If there was a more engaging app, then yeah, those are high stakes.
46:29 Harry Stebbings? Do you agree then with the notion that if you believe in AGI, you cannot be investing in SaaS apps?
46:35 Speaker 00 Yes, I absolutely believe that. SaaS as we know it, I think is over. I think it's completely over. I feel like quite a few SaaS apps were built at a time when software was relatively hard to build and complex to build. Imagine if you were building some customer support software, some customer support bot. To build a company like that, you would have had to hire some Stanford PhDs in NLP. You'll collect data for six months. You'll use like a support vector machine or a neural network that'll kind of sort of work. And then you'll deploy it and you'll grind away for a while. There is a significant amount of capital that needs to be invested to get an app like that to work well.
47:16 Speaker 00 So it made sense for many companies to not bother doing that if that's not their core business. Let me just use like some third party SaaS app. Now, many of these AI applications are incredibly easy to build on top of these LLMs. So I feel like most companies will start building custom software super easily. I mean, we help companies build some of these custom apps. The bar to create many of these apps significantly comes down. So that's one risk. Risk number one, companies do it themselves. Risk number two is you get sonic boomed by the foundation model companies.
47:51 Harry Stebbings? Sonic boom by, meaning they move into the apps layer and just create it themselves?
47:55 Speaker 00 Yes, like it could happen, right? Like the the models are becoming agentic. I mean, you you've seen many of these agents, right? It's fundamentally, it's about computer use agents. If the models get better and better there, it's possible like the the model is all you need. Imagine if you wanted the model to, let's say I'm doing some HR thing. Hey, update my medical benefits information to add, we have, we've just had a new daughter. We wanna update my medical information. If the model is agentic and it is sufficiently integrated into the database of the organization, you don't need anything else in the middle. So that's the second risk, the model's becoming more agentic. The third, and I worry about this a lot, I feel like a lot of our software was designed to be used by humans.
48:40 Speaker 00 Humans navigating a GUI and clicking around and doing things, I think that's gonna go away. I think of four pillars to superintelligence, it's multimodality, reasoning, tool use and coding. Multimodality is important because we humans interact in natural language, we talk, this video. The future might look like some type of ambient AI that you talk to that will just go and do things. And maybe it'll use the GUI of the current SaaS application as like an intermediate step. Or it'll use MCP and use tool calls and get what it needs. The GUI was like designed for a world where humans were using a keyboard and a mouse and clicking around and doing things. I think humans can do better things with their time than click around and.
49:20 Harry Stebbings? Actually, one big change that I have is I never actually type emails anymore. I use Whisperflow and it's it's so good in transcription that it's, I don't ever type emails ever.
49:28 Speaker 00 Yeah.
49:29 Harry Stebbings? Now the only trouble is everyone knows what I'm saying in my emails.
49:32 Speaker 00 The Whisperflow founders interned at Turing back in the day.
49:35 Harry Stebbings? No way.
49:36 Speaker 00 Yeah.
49:36 Harry Stebbings? So how do I feel about it? No, I I don't agree. Why? Because the average company today has between eighty and a hundred different SaaS products that they engage with. So one, just the multitude of how many they'd have to create, number one. Number two, maintaining them? You think they're gonna maintain eighty to a hundred? Oh my god, you're gonna have like just teams and teams of people doing maintenance and updates and debugging? I don't think so either. And that is for the technology savvy. Let's talk about every plumbing. Provider, law firm, accounting firm, restaurant that can barely use Wix and Squarespace, let alone build out their own CRM system and POS system. Not a fricking chance. And then we move to foundation model companies, moving into very vertically
50:22 Harry Stebbings? specific elements. You know, we're in a business that does AI for patent creation, updates and collaboration. SAM is not going there. SAM has health, you know, solving cancer, you know, energy utilization. I don't think SAM's touching patent creation and updating. And so I think for the more vertical you go, the more defensibility you have. And so I think for those reasons, SaaS has life. Mine is a very biased perspective because it's my job. Is that all wrong again? You're the master here, John. And so like, VC is a literally middleman.
50:57 Speaker 00 I'd say, Harry, like you have an interesting data set because you invest in a ton of startups. So I would be curious, looking at your sample of startups that you've invested in, and just tally how many SaaS apps they use today at every stage and see if that has changed, like post ChatGPT. My hypothesis is that today's companies use fewer SaaS apps and have fewer people.
51:22 Harry Stebbings? Do you think we have more or less software engineers in 10 years?
51:26 Speaker 00 More. I think the definition of a software engineer will change. A Stanford doctor who's in oncology, who has an idea for like some cancer detection type app. Now that person will be able to create like a very simple version of an app that somebody could check by themselves and send, do like a home diagnosis. I think there'll be more software engineers because if you define a software engineer as somebody who's capable of building a software product to solve a real problem, that pool of builders is going to expand way beyond people who've graduated with like a four-year computer science degree.
52:01 Harry Stebbings? So we have more software engineers creating more software. And the problem then becomes discovery, you know. How do we solve the discovery problem in a world of infinite software?
52:11 Speaker 00 You might have an agent for yourself that's talking to other agents on the internet. You might have an agent of yourself that's like discovering startups to invest in. And while you and I are chatting, I don't know, you might be having like a million conversations with entrepreneurs from all over the world who today you're you're kind of constrained by space and time, but in the future you'll only be compute and data constrained.
52:32 Harry Stebbings? Do we lose the phone as the interface to this world? We obviously see Sam and Johnny, I've there's rumors of pendants and some hardware devices. I'm not asking you to comment on that. I'm just saying, does the phone still remain the primary interface and design device?
52:47 Speaker 00 I think we'll have some type of a device that will carry that's always on and processing multimodal tokens. For example, as I'm talking to you, if I were to envision my perfect device, it would be something that has it should have cameras. So maybe it's a wearable as a glass or something that I'm having on me that's processing visual input because I want to be able to read your body language. I might have like an AirPod like thing in my ear that's whispering to me that maybe says, Jonathan, as you were talking about multimodality, Harry seemed less interested. His his body cues suggest that he was losing interest. But when we were talking about AR, he poked up. So those types of feedback and cues I think would be good.
53:30 Speaker 00 So I envision a device that I think of it in terms of sensors and effectors. In terms of sensors, like obviously it has to be listening to stuff. It has to be seeing stuff. But in terms of effectors, it'll probably also be speaking in my ear. Ideally, it should be something that you can talk to and have it do things later. For example, I might say, remind me to follow up with Harry on that idea for using Turing to automate clip generation. So it has to like remember that and come back later. So I do think there'll be all sorts of new devices and glasses, hearing like these AirPod type devices seem obvious. They could be like, have you, do you remember this device that was called the meeting owl?
54:12 Harry Stebbings? No.
54:13 Speaker 00 Like during like the COVID era, like one of the tools that sort of spiked was like this. It was basically like a speakerphone for like having better distributed team Zoom meetings.
54:23 Harry Stebbings? Okay.
54:24 Speaker 00 When somebody's talking, it would focus on them like with like a camera. It was also a decent speaker, so I can imagine devices like that. So it's hard, it's hard to predict. But the thing that I almost feel confident about is that the phone will look so different. I mean, when we think of our smartphone, it's basically a computer with like a phone app in it, right? Like the phone app is like the least interesting part of the phone. And I think even for for an AI device, it'll probably have some phone app in it. But everything else I feel like will be magical. Like, I feel like I would definitely benefit from a device that's constantly listening to everything that I'm listening to, constantly processing all the video, audio input that I'm processing.
55:05 Speaker 00 Something that's paging things to memory, like maybe it'll like write things down and be able to look it up later. I see it almost like an extension of my brain.
55:12 Harry Stebbings? Before we move into the quickfire round, I do just have to ask, what does your market and the data provisioning market look like in 10 years? I always try and think about like market composition and dynamics. Is it a winner take all? Is it a very fragmented? Is it a three or four? What does that look like?
55:30 Speaker 00 The market will reward players with research depth because the pace of AI research is so rapid. Like all these RL environments like this spiked in like the last 12 months after O1 came out in December and DeepSeek came out in Jan. So now it's like in addition to imitation learning, we are in this reinforcement learning regime. One year later, it could be something totally different. So I think the market will reward a company with research DNA. And it'll reward a company that can move fast and adapt very quickly. Because this is, I mean, when you're serving.
56:05 Harry Stebbings? Do you think this is a monopoly market or do you think there will be many winners?
56:08 Speaker 00 I think there'll be a few winners. I do think for the labs, it helps them to have a few partners for resiliency. I imagine also for price competitiveness. I think there'll be a few winners. In the realm of robotics and embodied AI, we are still very early. At Turing, we are scaling up on the robotics side as well in terms of data that we generate. But there's so much data that's missing that the models need to see that they haven't seen yet. I can totally imagine some newer companies also coming up that don't exist today.
56:37 Harry Stebbings? If you were to invest in companies in your space, where would you invest?
56:42 Speaker 00 Probably in robotics or embodied AI. The vertical stuff like, I mean, we are scaling up pretty massively in generating data for different verticals. I don't see that as like a big white space. But I think everybody is relatively early with robotics. And robotics is such a vast realm that there could be interesting things to do there. One way I see the space, again, think of like these three dimensions. The first dimension being the type of intelligence that you're baking into the models. That could be in coding, in STEM, in functional expertise like sales, marketing, software engineering, or vertical expertise like healthcare, legal, finance, et cetera. So the first dimension is the type of intelligence.
57:23 Speaker 00 I do a cross-product of that with the modality, audio, video, image, computer use. So that's multimodality. That's the second dimension. The third dimension is multilinguality, like different languages. The fourth dimension is different learning paradigms, like imitation learning, reinforcement learning, pre-training as well, which is unsupervised learning. All of those may require different platforms to be built. Like, we've had to adapt our platform for imitation learning, for reinforcement learning, for multimodality. So I feel like in this matrix, there's like all sorts of new opportunities that could emerge. And I only listed like the digital intelligence. I didn't talk about physical intelligence. So I think robotics is like wide open. I mean, the kind of data that a robot that is in someone's home is totally different
58:08 Speaker 00 from like a robot that's doing things in a factory and humanoid versus non-humanoid robots and.
58:13 Harry Stebbings? I could talk to you all day. I do want to move into a quickfire answer. I say a short statement, you give me your immediate thoughts. What's one widely held belief about AI that you think is wrong?
58:23 Speaker 00 I don't think we'll see rapid takeoff. I think we'll see incremental, continuous improvement in AI. I actually think this is good for the world because if what we believe happens, which is all types of digital knowledge work gets automated, I think humanity needs time to prepare its workforce. I think we could use the extra time to upskill humans, to rethink education, to make sure there isn't massive job displacement. I also think. In the steady, continuous improvement in AI models, there'll be value realized every step of the way. Unlike self-driving cars, I feel like people have this wrong model for AI that comes from self-driving cars,
59:05 Speaker 00 where you get it ninety-nine percent of the way accurate and you, if you can't solve the last one percent, they're not useful. AGI is not like that. I think when we automate the job of an underwriter or a claims processor or a CEO, there's incremental value that's unlocked for every percentage improvement in the models becoming more reliable. I believe in slow and steady takeoff, and that's actually going to be great for the world. You mentioned DeepSeek a couple of times. Do you think we underestimate China? It depends on who you ask. The folks that I work closely with don't underestimate China. I think it's very impressive, like the progress that they've made in open source with DeepSeek, Kimi K two, Qwen.
59:45 Speaker 00 These models are state of the art. So, no, I don't think, at least among like the frontier AI circles that I'm in, I think there is a clear realization of like how close they are. The world seems to be moving to closed models. Is that good or bad? I think it depends on the application. Firstly, in enterprises, it's often a mix between closed and open models. We do see demand from enterprises that want either. The closed models often are easier to get started with, but there are some cases where enterprises prefer open models for cost, customizability.
1:00:26 Speaker 00 And I'm talking about the small language model regime between the half a billion parameters to ten billion parameters. I worry a little about frontier models. I feel like for frontier models, there is some value in keeping some of the technology closed, just because of how powerful they are. And I feel like the US labs are extremely responsible and safety-conscious in how they think about training these models, deploying them. Do you think Elon is, you mentioned reading his book earlier, Elon is often chastised for his lack of care?
1:01:01 Harry Stebbings? Around some of the training elements. Do you think he is and do you think actually he'll benefit from not having that guardrail?
1:01:08 Speaker 00 I think Elon is also like he cares a lot about humanity too. Like the at least if you in in his book, like one of the things I recall reading is his motivation for getting into AI was that he wanted an AI that was speciest and loves humanity. That was one of his reasons to get into it. Everything I see about like the the Grok team, I feel like their goals are much like any of the frontier labs, like quite noble in terms of having this powerful AI that can help humanity, understand the universe, solve some of our biggest problems.
1:01:42 Harry Stebbings? What did you believe that you now no longer believe?
1:01:46 Speaker 00 I used to believe that to build a enduring valuable company, you hire a strong exec team, operate with a lot of leverage, hire great people and get out of the way. I used to believe that. Now I believe you hire great people and work really closely with them and their directs and their directs and their directs. Get as close to the ground as you can, where ground truth usually exists with the customers. The next step to customers is the engineers writing code and the salespeople talking to your customers. Now I believe in being basically, I used to like, for lack of a better word, follow the org chart a little bit.
1:02:26 Speaker 00 And this was also part of one of my learnings from Elon's biography is that he was so hands on. Like he would be like walking the factory floor and asking an engineer why this door in the Model Three has three bolts instead of maybe two. It is a different way to operate where you're in the details of the most important things that matter. Completely working in like a flat structure, operating as close to the to the ground truth as you can. I feel like in the early days of starting Turing, like now that I think about it, I may have had a subconscious desire to be liked. I think I must have. Now I don't. Now I just think about just doing things that would solve our customers' problem the best.
1:03:11 Harry Stebbings? What was the most unpopular decision you've taken with Turing?
1:03:14 Speaker 00 Turing is switching from a distributed team to a hub and spoke model. So we are now working from an office in San Francisco. And we've recently opened an office in Palo Alto. We're going to be opening an office in London as well. For some people, like, that wasn't very popular. Some of them left and yeah.
1:03:33 Harry Stebbings? We like in person. We're big fans of in person here. Final one. When you look forward to the next decade, what are you most excited for? So for me, my mother has MS. I think we'll have some pretty groundbreaking breakthroughs in MS drug discovery that we haven't had for ever. That excites me.
1:03:51 Speaker 00 I'm excited about AI making new discoveries, automating AI research itself. To get to a point where AI is in some self-improvement loop. So that we could get to superintelligence faster. So automating AI research and getting AI to the point of making new breakthrough discoveries. I'm excited by that. And I've always been fascinated by AI as like this exoskeleton that makes you a lot more productive. Have you watched the Iron Man movies, Harry?
1:04:22 Harry Stebbings? Yeah.
1:04:23 Speaker 00 In the early Iron Man movies, like, he's wearing like the suit and the suit is obviously giving him superpowers, right? And then in the later ones, the the suit is quote unquote agentic. Where like he has these drone suits, right? Like where there's like a army of his suits that go off and do things. I'm excited about a future like that where every human on the planet has access to agentic AIs that help them amplify their fullest potential. Today, Harry might have a hundred ideas, but Harry's able to do maybe two of them really well. I like a future where Harry can do the remaining ninety eight. And I like that for like the seven billion humans on earth.
1:05:02 Harry Stebbings? Jonathan, I I love conversations which are very natural and free flowing. You can tell that I don't really pay much attention to the schedule, but you've been fantastic. So thank you so much for joining me.
1:05:12 Speaker 00 Thank you, Harry, for having me.
1:05:15 Harry Stebbings? But before we leave you today, are you drowning in AI tools? ChatGPT for writing, Notion for docs, Gmail for email, Slack for comms? And you're constantly copy pasting between them all, losing context and losing time? This is the AI productivity tax and it's killing your output. At 20VC, we're all about speed of execution and Superhuman is the AI productivity suite that gives you superpowers everywhere you work. With the intelligence of Grammarly, mail and coder built in, you can get things done faster and collaborate seamlessly. Finally, AI that works where you work, however you work. Superhuman gets you from day one with zero learning curve
1:05:59 Harry Stebbings? and it's personalised to sound like you at your best, not like everyone else using generic AI. Get AI that works where you work, unlock your superhuman potential. Learn more at superhuman.com/podcast. That's superhuman.com/podcast. And just like Superhuman gives you superpowers in your inbox, Vanta gives your company superpowers in security and compliance. Customer trust can make or break your business and the more your business grows, the more complex your security and compliance tools get. It can turn into chaos and chaos isn't a security strategy. That's where Vanta comes in. Think of Vanta as your always-on AI-powered security expert who scales with you.
1:06:42 Harry Stebbings? Vanta automates compliance, continuously monitors your controls and gives you a single source of truth for compliance and risk. So whether you're a fast-growing startup like Cursor or an enterprise like Snowflake, Vanta fits easily into your existing workflows so you can keep growing a company your customers can trust. My listeners can get $1,000 off Vanta by going to vanta.com/20vc. That's V-A-N-T-A dot com forward slash 20vc, 20vc, for $1,000 off Vanta. And like Vanta builds trust with automated security, AngelList drives momentum by helping you find and fund breakout startups.
1:07:24 Harry Stebbings? It's 20VC, you know we have a really frickin high bar. Well, Angel List is the modern platform used by the best-in-class venture funds, where over 40% of top endowments and banks are LPs. Their customers include a top-five venture firm, 20VC, and they now have, check this out, $171 billion of assets on the platform. They combine an all-in-one software platform with a dedicated service team that moves as fast as you do. One manager said this awesome quote, "Angel List feels like an extension of my fund." Another said, "Angel List gives me total peace of mind, the attention to detail, lightning-fast response time, and just real sense of ownership from the team are exactly what I need to stop worrying about back-office ops."
1:08:06 Harry Stebbings? So if you're starting a new fund, don't be a moron, just use Angel List, they're incredible. Head over to angellist.com/20vc to learn more.
Previous 20Sales: John McMahon on How to Hire, Train & Retain the Best Sales Reps | How Sales Changes in a World of AI | Sales Lessons from Snowflake and MongoDB | How to Create and Drive a Sales Process with Urgency Next 20VC: Thrive & OpenAI Partnership | Eventbrite Acquired for $500M | Databricks Raising $5BN at $134BN Valuation: Cheap or Not? | Why SaaS is Like Japan and The TAM Trap in Software Page updated September 29, 2026 · transcript and summary are machine-generated
