Pod Insights Y Combinator Podcast From Idea to $650M Exit: Lessons in Building AI Startups About this episode Jake Heller is the co-founder & CEO of Casetext, the AI legal startup behind CoCounsel, which was acquired by Thomson Reuters for $650 million. In his talk at AI Startup School on June 17th, 2025, he shared how his team did it—from picking the right idea to building AI products that actually work—and how founders can turn a cool demo into a reliable tool used by real customers.
Transcript 0:00 Speaker 00 What we're going to talk about today is how my company built an AI app that was so good, we're able to bring it to an exit for $650 million and how you can do that too. All right. So really, we're talking about three big ideas today. The first is what ideas to pick? How do you decide what to pursue? Second is how you actually build it. And third, and honestly, often overlooked, is how you take that thing that you built and market and sell it successfully in the market. Before we dive into this, a little bit about me, so you know who's talking to you. I grew up a coder. I've been building stuffs as long as I can remember. It's probably the same as basically everybody here.
0:45 Speaker 00 Bit of a side quest for me, but I fell in love with law and policy and I became a lawyer. And I had a pretty conventional, though brief, legal career. Law school, clerkship, you know, big law firm, et cetera. I think like anybody who builds stuff and then goes to one of these old professions like law or accounting or finance or whatever, the first thing you find out is I cannot believe that they are doing it this way. And so I immediately left that and founded a company called CaseText in 2013, when I think a lot of you are about turning eight. And maybe as a side note, that's about how long it takes sometimes for these companies to be successful. So I know you're, you know, 18, 19, 20, 21, 22, whatever you're old right now. Be ready to sign up for one of the most amazing adventures of your life when you start a startup,
1:30 Speaker 00 and it takes time. At CaseText, we've been focused for, you know, the vast majority of our experience on a deep conviction that AI, when applied to law, can make a huge difference. And by the way, it wasn't even called AI when we started focusing on it. It was called natural language processing, maybe machine learning. But one of our AI researchers, who is here today, Javend, saw, woo, saw an early application. As soon as the BERT paper came out, attention's all you need, et cetera. It's like seven years ago of how AI technology could apply to making lawyers' lives better. For example, making search a lot better. Because we were so focused on large language models and were researching deeply in this space, we got really early access to GPT-4.
2:17 Speaker 00 Like summer 2022, we were like $20 million in revenue. We were doing great. I had like 100 people and we stopped everything that we were doing and said, we're going to build something totally new based on this new technology. And that became a product called Co-Counsel, which was the first ever, and I think still the best AI assistant for lawyers. For reasons I'll go into the rest of this talk, we were acquired by Thomson Reuters not about two years ago for $650 million in cash. By the way, that feels like a big number, but I think for a lot of folks in this room, you're going to look back at this talk and be like, I can't believe that was a big number back then. You guys are going to be able to build things that are so much more valuable. I really believe that. And I think that's because the what AI is going to unlock for all of you is the ability to build amazing stuff in this for this world.
3:01 Speaker 00 So, okay, how do you pick an idea? Like, how do you know what to work on? It's actually one of the hardest and most consequential problems. Since like the beginning of YC, you know, they've had this saying, make something people want. And the reason they had that saying is because it's genuinely difficult to know what people want, especially in like the old world of building software. You kind of like have to build something, get it in users' hands and try and fail a lot of different times. And you just hope that it's something that people actually want to use. So that's why the saying for why Commodore is make something people want. I actually think it just got a lot easier because what do people want? Well, what do people want? For example, things they're paying for right now. People are currently paying people to do tasks, right?
3:46 Speaker 00 In this case, it's one of very unhappy like customer support people or something like that. But we already know what people want because they're paying people to do it. This includes a lot of work like customer support or insurance adjusters or paralegals or in, you know, things you do in your personal life, like personal trainers or executive assistants or whatever. That is what people want. And so the, the problem of choosing what people want just got a lot easier because now you just have to look, what are people paying other people to do? For a lot of those problems, either, you know, traditional AI like LLMs can solve many of the problems that people work on right now. And if not that, then robotics can solve a lot of things that people are working on in the physical world. And what I think you're going to see as you decide what you're going to build, you first
4:28 Speaker 00 pick an area to target, it really kind of falls under three different categories. One is like assistance, where say a professional needs help accomplishing a task. That's what we built with Co-Counsel. Lawyers need a lot of help, reading a lot of documents, doing research, reviewing contracts, marking them up, making red lines, sending them to opposing counsel. So that's one big category, is assisting people doing their work. The second big category is just replacing the work altogether. People currently hire lawyers. What if we just became a law firm powered by AI? People currently hire accountants and find financial experts and physical therapists and, and, and, you know, people to fold your laundry, whatever it may be, right? You can just replace that task using AI.
5:11 Speaker 00 And finally, the third category is you can do things that were previously unthinkable, right? Like, for example, at law firms, they would have hundreds of millions of documents. And they would never think in a million years, I should have people read over every single document and categorize it in certain ways and summarize it and index it, et cetera. It'd be insane, right? It'd cost them millions and millions and millions of dollars. But now that AI is here, you can have thousands of instances of Gemini 2.0 flash or whatever, read over every document. The previously unthinkable is now thinkable. These are basically the three categories of ideas to choose. And what I think is incredible about this is the amount of money to be made with these
5:54 Speaker 00 new kind of categories each has gone way up. It used to be that what's called the total addressable market, which is basically how much money you can make from your product, was the number of like professionals, for example, number of seats you can sell, times the dollars, like $20 per month or whatever, right? And by the way, a lot of many billion dollar companies were built selling seats to X number of professionals. But today, the actual amount of money that we already know people and companies are willing to spend Is the combined salaries of all the people they're currently paying to do the job. And that number is like a thousand X bigger. You pay $20 a month to solve a problem.
6:37 Speaker 00 For example, you know, pay a typical SaaS kind of subscription. But you might pay five or 10 or even $20,000 a month to certain professionals to solve problems for you. So the amount of money that you can make with your new applications with AI has gone up by a factor of 10, 100, or even 1,000 compared to what it used to be. I want to take a quick moment because it might sound like pretty dystopian. Like we're talking about taking all these salaries and these become, you know, your addressable market. I think it's kind of the opposite. I think it's beautiful. I think the future is beautiful for two reasons. The first is that you're going to unlock a future. When you replace or substantially assist certain jobs, like people used to, Sam Altman wrote about this in a recent essay.
7:22 Speaker 00 People used to have a job called lamplighters where we didn't have like, you know, electricity and lights. So people go around with a like matchstick, you know, lighting all the lamps at night on and then turning them off at night by putting out the candles. Right? That's what things used to be. And we couldn't even imagine the kind of stuff we're doing now because that's what we were stuck doing in the past. So you're going to unlock a future that we can't even imagine today. When we, you know, move past the roles that we're currently doing right now. It'll feel antiquated 10 or 15 or 100 years from now to do the kind of things we're doing today. Because you're going to help us move past that. But as importantly, what I think some people don't think about with this stuff, which is very true,
8:06 Speaker 00 is you're going to democratize access to things that were, used to be really, really hard or very expensive. In the field we worked in in law, over 85% of people who are low income don't get access to legal services. It takes way too long and it's way too expensive, working with human lawyers. Right? But if you could help make lawyers 100 X faster and 10 X cheaper, or frankly, just provide those services yourself as a new law firm powered by AI, then all of a sudden saying, where lawyers have to turn away clients because they did not have enough money, you can now say yes. And that applies everywhere. Everybody should get the world's best financial assistant. Everyone in the world should get the best executive or personal assistant.
8:49 Speaker 00 Everyone in the world can already have the best coding assistant in tools like Curse, Cursor and Wind Surf, et cetera, right? I do think that despite the fact that I'm telling you how to pick an idea is you should potentially replace jobs, I think you're going to do something really amazing for the vast majority of consumers and enterprises by unlocking a better future and by democratizing access to things that used to be only for the very wealthy. Okay, so that's how to pick an idea. Pick a job, replace, assist, or do the unthinkable, those previously unthinkable, and build a better future. But how do you actually build this stuff? I'm going to give you a quick outline of how we built it. What's kind of nuts to me is everything I'm going to say right now may sound very simple and commonsensical and maybe even obvious, but the craziest shit is nobody's
9:34 Speaker 00 doing it. Like, nobody's picking ideas the way that I recommended in terms of picking job categories. There's very, very few companies out there doing that. And even fewer companies are doing what I hope will look like pretty obvious and simple things to building reliable AI. I put a reliable in underscore for what it's worth because that's going to be the key for many circumstances in terms of getting from a cool demo, as Andrew was saying earlier today, something that actually works in practice. Here's like four quick points about how to actually build this thing. The first is, think about like making an AI assistant or an AI replacement for, say, a profession. Ask yourself like, what do people actually do?
10:19 Speaker 00 What does a professional in this field actually do? What does a personal trainer or fitness coach do? If that's the app you're deciding to build. What does a financial assistant do or financial analyst do? And be like super specific. I'm going to say this a few times, but it is really helpful to actually know this answer, not like make it up. It was helpful for us that I was a lawyer, my co-founders were lawyers, 30 to 40% of my company, even the coders were lawyers, because we actually lived it.
10:48 Speaker 00 That may not be the case for you. Just go be like an undercover agent somewhere. Really learn what happens at these companies. Right? What do these people do? Other way to do it, by the way, is you might be the tech talent. You may find yourself a co-founder who has some deep expertise in a field. But whatever way you get there, you know, find out what are the specific things that people do that you can assist or replace. And then ask yourself this question. How would the best person in that field do this? If they have like unlimited time and unlimited resources, like a thousand AIs that can all work in, you know, simultaneously to accomplish this task. Right? How would the best person do this? And work backwards from there. Right?
11:32 Speaker 00 What are the actual steps that somebody might take to accomplish a task? For just give you an example from our legal field. We did a version of deep research two and a half years ago.
11:45 Speaker 00 As soon as we got access to GPT-4, it was like the first thing that we did. And we asked like, what was the, what was the best lawyer going to do? If given this research question? It wasn't like just generally research. Like, what does it even mean? It broke it down to steps. Okay, first, you know, they get a request for this research project. And they say, okay, well, I need to understand what this really means. They might ask clarifying questions, quite like deep research today, if you've used it. And then they might make a research plan. They, they might execute dozens of searches. But that might be back hundreds of different results. They'll read every single one of them very carefully. Kick out the stuff that's not relevant because search results are sometimes have irrelevant stuff. Bring in the stuff that is relevant. Make notes about what they're seeing. Right? Why is this relevant? Why is this not relevant? Where does this fit into my answer?
12:30 Speaker 00 And then based on all of that, put together, put it all together in an essay. And then maybe even have a step at the end where you check the essay to make sure it's accurate. And really, you know, actually refers to the right resources, et cetera, et cetera, et cetera. These are the kind of steps that a real professional might do when doing research. So write them down. Now we turn to code. Most of these steps for the kinds of things you'll be doing end up being prompts, one or many prompts, right? One prompt might be read the legal opinion and decide on a scale of zero to seven, how relevant is it to the question that's being asked? One prompt might be, given all these notes I've taken and all the cases I've read so far, write the essay.
13:12 Speaker 00 One prompt might be like, here's a footnote in the essay. Here's the original resource. Is this thing accurately cited or not? The reason why many of them are prompts is because they're the kinds of things that would once require human level intelligence, but now you're injecting it into like a software application. So now you need to do the work of turning it into a great prompt, we'll talk about in one second, to actually do that human level intelligence. By the way, if you can get away with it not being a prompt, if it's like deterministic or it's like a math calculation or something like that, that's better. Prompts are slow and expensive. Tokens are still expensive. So when you're breaking down these steps, some of these things might just be good old
13:57 Speaker 00 software engineering, right? Do that when you can. And then here you make a decision. When you find out how the best person would approach this, if it's a pretty deterministic, like every single time they always do this task, they always follow the same five steps, simple, make it a workflow, right? It's actually the easiest outcome for you. And to be honest, a lot of the stuff that we built while building Co-Counsel was exactly like this. Every time you do this task, you're basically going to take the same six or seven steps. And you don't need to have frankly like fucking Langchain or whatever, just Python code. This function, then output of this function goes to this function, output of this function to this function, boom, you're done, right? Simple. Sometimes it's not so simple. Sometimes how well expert would approach the problem really depends on the circumstances.
14:42 Speaker 00 Maybe they need to make a very different kind of research plan, pull from different resources, run different kinds of searches, read different kinds of documents, whatever it may be that you're doing, right? That's where you get to something that's a little bit more agentic. That's harder to make sure it's good, but maybe what you have to do, right? Underscore this again, in doing all of this, having some form of domain expertise, somebody who knows what they're talking about here, which by the way, you can also acquire it by talking to a lot of people, there are lots of different ways to get here, but don't do it, don't fly blind. Don't assume this is the way that all government employees in this field do X. Really no. Okay, so that's the basic way you can build these AI capabilities that start to round out, and that's it, right? Simple.
15:26 Speaker 00 The hard part, frankly, isn't building it. The hard part's getting it right. Like, how do you know the research was done well? How you know it read the document right? How do you know it edited, you know, it did the insurance adjustment correctly? How do you know it made a correct prediction about whether to buy or sell a sock? Or whatever it is that you're doing. This is where evaluations play a very, very, very large part. And this is the thing that I see most people not doing because they build like demo level stuff that frankly is like 60 to 70% accurate. And if we're being honest, you can probably raise a pretty good round of capital by showing your cool demo to VC partners. And you can even possibly sign on your first few customers with a cool demo as a pilot program, right? But then it doesn't work in practice.
16:10 Speaker 00 And so all that excitement and VC capital raised and pilot program excitement, et cetera, falls apart if you can't make something that actually works in practice. And making something that works in practice is really hard because LLMs, like people, you know, you don't have your coffee that morning. You wake up on the wrong side of the bed. You might just output the wrong stuff for prompts. I'm sure you've all seen this before. Even if you just use ChatGPT, you sometimes probably have been blown away with its brilliance at times. And then other times shocked by how incredibly wrong it was about code or, you know, some informational lookup or just hallucinating when George Washington's birthday was or whatever it is, right?
16:55 Speaker 00 So how do you deal with that? I'll tell you how we dealt with it. This is not the whole answer, but a big part is evaluations.
17:01 Speaker 01 YC's next batch is now taking applications. Got a startup in you? Apply at ycombinator.com/apply. It's never too early and filling out the app will level up your idea. Okay, back to the video.
17:14 Speaker 00 This all begins again from domain expertise, which is like, what does good look like? What does it mean to do this task super, super well?
17:24 Speaker 00 If you're doing research, what is the, you know, for X given question, what is the right answer? What must, what must the right answer include? For X document, and you're asking a question at the document, what must it pull out of that document? What pages should define that information? What does good look like? This is true of the overall task, like, "Complete this research for me," but also each microtask necessary to complete the overall task. Like, which, which search queries are good search queries versus bad search queries? Here again, not sounding like a broken record, but it's good to know what, like, actually, actual professionals would say about this, right? So what does good look like? And then those become your evals. My favorite thing to do when I'm writing evals for things that are like,
18:08 Speaker 00 you know, when, when possible, is to turn it into like a very objectively gradable answer. For example, have the AI just output true or false, or a number between 0 and 7, or whatever. Because then it's really easy to eval. For the given document and given question, the answer should be 6 on a scale of 0 to 7. That's how relevant it is. It's not a 7, not a 5, it's a 6. And if you have that, then you can set up an eval in a framework. I like Promptfoo. I don't know if you guys use that. It's like open source, runs on, on your command line. There are many frameworks out there that you can use to, you know, put together these evaluations. Doesn't really matter. At the end of the day, it's like, for this input and this prompt, the answer should be 6.
18:53 Speaker 00 Make like a dozen. Try to match what your customers are actually going to throw at your program, right? Make a dozen, and then try to get it perfect on a dozen. Then get to 50, then get to 100. And keep on tweaking the prompt until it actually passes all the tests you keep on throwing at it. If you're doing really good about this, have a holdout set. And don't, you know, look at those while you're, while you're writing your prompts. Make sure it also works on those. You're not just, just fine tuning the prompt just for your evals, right? What you'll find, without any, I use the word fine tuning, without any like technical fine tuning, you can go so far with just prompting. If you're being really careful about this, you will find that the AI gets things wrong predictably. You're ambiguous
19:35 Speaker 00 as part of your prompts. You're not giving it clear instructions about doing one thing, or maybe it just constantly fails in a certain direction, and you have to give it, you know, prompting instructions to pull it back from making this kind of error. You give it examples, right, to guide it away from certain classes of errors. But it's not like gonna be a surprise why or how AI fails. Once you start prompting, you'll start to see patterns that you can prompt around and give instructions around. And what I like to say is like, the biggest qualification for success here is whether you or whoever's working on the prompts in your company is willing to spend two weeks sleeplessly working on a single prompt to try to pass these evals. If you're willing to do that, you're in a really good place,
20:19 Speaker 00 right? It just, it just takes such a grind, because the thing is, you're gonna do these evals, and at first you're gonna pass like 60% of the time. And at this point, most people just fucking give up. They're like, AI just can't do this task, right? They're like, I just can't. I'm not gonna do it. And then you'll spend a night prompting, and you're gonna be at 61%. You're like, oh my god. The next group of people will give up at this point. What I'm here to tell you is that if you spend like solid two weeks prompting and adding more evals and prompting, adding more evals, and tweaking your prompt, and tweaking your prompt, tweaking your prompt, you're gonna get to something that passes like 97% of the time, and the 3% is kind of explainable. It's like a human would, it's like a judgment call almost. Humans make similar kind of judgment calls. Once you're there, you're gonna feel pretty good about how this might interact in, in, in a production.
21:03 Speaker 00 What I recommend is like pre-production, maybe in like beta, get to 100, you know, tests per prompt and 100 tests for the overall task. If you're passing like 99 out of 100, again, you should feel pretty good about where you are, right? So that's just a rough guide. If you can beat 1,000, that's 10 times better, do that. But it's hard. It's actually really hard to come up with great evals. So I'd recommend you just do at least 100, go to beta, and put it in customers' hands. And set the expectation, by the way, this is not yet perfect, that's why you're in a beta. And then you listen and learn. Every time a customer complains, either you have their data because that's how your app is set up, or you ask them like, hey, can you share that document and that question you asked? See why it failed?
21:47 Speaker 00 That's a new test. We've added much more evals at this point from real things that happen to real customers than the ones we came up with in the lab. And that's going to, your customers are going to do the dumbest shit with their app. Okay, they're going to do such dumb things that you'd not predict. But that's what customers really do. If you've ever seen like a real person's Google queries, they're barely legible. You know, and I'm assuming the same thing is true of ChatGPT. They see a bunch of stuff. Like your prompts probably look pretty smart. Most people are like, burrito, me, near, how, ouch, or whatever. Like, what do you do with that? Right? But you have to try to bring back a great result and determine what they're actually trying to say with these ridiculous prompts. So do it. Like those become your real tests and just keep iterating.
22:32 Speaker 00 This is not a static thing. New models will come out. Try the new models. Prompt Fu and other frameworks make this really easy. Add a new model, it'll compute how well it does against your prompts so far. Keep tweaking your prompts. Sometimes the addition or subtraction of a single word might move you up a single percent, but that's a very big deal if you're working in a field like finance, medicine, law, where single percentage increases in accuracy really matter to the customers you're serving. Right? Keep iterating, never stop. There should be a new GitHub pull request like every other day or every day on your prompts. And I'm telling you, if you just do those two last slides, you know, how do the professionals really do it? Break it down into steps. Each step basically becomes a prompt or a piece of code. And then you test each step, test the whole workflow altogether.
23:17 Speaker 00 If you just do these two things, you'll be like 90% of your way there to building a better AI app than what most of the crap is out there. Right? Because most people never eval and they never take the time to figure out how professionals really do the job. And so they make these kind of flashy demos on Twitter. They maybe even raise capital and they may even be some of your like your heroes for a minute. But be careful who chooses your heroes. The real people are behind the scenes quietly building quietly making their stuff better every single day. You just do these two slides, you're going to be 90% of the way there. And better than most of the things that are out there. That's the craziest part. Okay, now the hardest part, honestly. And the part that frankly, we are still trying to figure out. Post-exit, you know, at a multi-billion dollar company.
24:01 Speaker 00 It's still going to be really, really, really hard. And I'm going to give some tips about. Marketing and selling AI apps in this new kind of world where you're maybe replacing or assisting a job. Things that we've learned along the way. But the first thing I'll say. This is a little bit counter to what I think is out there in a lot of the VC kind of. A lot of people like say like the most important thing is sales and marketing. A lot of people really, really think that when you guys raise series As and series Bs. You'll have people on your board. You say product doesn't really matter that much if you're really good at marketing and selling. And they've seen some examples of this working out like really well. I think it's fucking bullshit. We for 10 years, we had an okay product. At first, we went through different marketing and sales leaders, some of them super. You know, well qualified, etc. And they did okay.
24:46 Speaker 00 When we had an awesome product, all of a sudden. People were referring us by word of mouth. News was coming to us because we're doing something genuinely new and interesting. Right. And that and word of mouth and news is free marketing. People coming to you like we had salespeople. Because we had salespeople from our older product that wasn't as good as the new one that we came out with with. You know, based on L L Ms. And I will tell you those salespeople became like order takers. So the most important thing you could do for marketing and sales. Is to build a fucking amazing product and then making sure the world knows about it somehow. Obviously can't just like build it and not show anybody. Tree falling in the woods, nobody hears it, it's not going to do anything. But I do think that the quality of product matters so much more than your series A and B.
25:30 Speaker 00 Investors will say, so when you guys have those lame VCs on your board, you can think back to this talk and pushback. All right. But it's still important, it's still important to market and sell. I have just three pieces of ice here. The first thing is. You might not be selling traditional software anymore. Think about how you're going to package and sell it. The companies I'm most excited about right now are taking real services, like for example, reviewing contracts for a company. And they're just doing it. They're like doing the full service. Maybe there's a human in the loop. And this would usually cost somebody $1,000 per contract to review. If they went with a traditional law firm. They're charging $500 per contract. All right, again, for context, a lot of the tools you guys use right now, probably 20 bucks a month.
26:13 Speaker 00 $20 per month versus $500 per contract. We're talking about extreme step ups in price. Price it according to the value you're selling it. Don't short come yourself. It's maybe a little in conflict with what I just said, but also listen to your customers for how they want to pay. Just ask them, how would you rather pay for this? I'll tell you what we found out. We were thinking about a per usage pricing like this review, viewing contract company. And that that may work in some cases where they prefer to pay that way. That might work. But when we asked our customers, they said, listen, I'd rather pay more, but make it like consistent throughout the year than potentially pay less and pay per use. So our customers wanted to pay $6,000 per seat. They want to proceed and they want to pay $6,000 for the proceed. 500 bucks a month. Fine.
26:57 Speaker 00 It's a situation where our customers wanted predictable budgeting. Give it to them, right? Listen to your customers. The third thing to really think about when you're marketed and selling is all this AI stuff is new and scary. These big companies even, they want to dip their toes in the water. They want to try new things. Their CEO is like sitting on a board of people at a Fortune 500 company. The whole board is like, what are you doing about AI? And so their CEO is going to this company of like 20,000 people. What are we doing about AI? And they're like, I don't know. I'm trying like Greg's product. Okay, they want to, they want to try your product. But there's also this trust gap. Because they used to do this thing by asking people and they can fire people. They can train people, they can coach people.
27:41 Speaker 00 Like people are not perfect, but they're used to them. They don't know, they're not used to using your product yet. They have like no idea what to expect. So how do you build trust? Some really smart companies are doing like head to head comparisons. Keep your law firm. And then use our thing side by side and then compare. How fast are we? How good were we? How different were the results? Keep your accountant, use our AI accountancy and then compare. Like, how different, how off are we in our accounting, our tax accounting, or whatever it is? Offer that. That's a great way to build trust. Compare it against people. Do studies. Do pilots. There are so many ways that you can do this, but think, think in your head, how do I build trust with my customer?
28:23 Speaker 00 And finally, the sale does not end when they've written the check and definitely not when they started a pilot. What I'm seeing right now is like an angel investor in this kind of post-exit world for, for me. Is there a lot of companies like our ARR is $10 million and you dig under the surface and it's like, oh yeah, we have a pilot for like six months and they're paying us a lot of money for that pilot. A lot of those pilots are not converting to real revenue and there's going to be a mass extinction event. As a lot of pilot revenue, it's like instead of ARR, it's like PRR, like pilot recurring revenue or something. They're not even recurring, just pilot revenue, I guess. Like, is this not going to convert into real money? And it's a real danger, I'd say, for startups right now. Even ones that are reporting super high numbers in terms of revenue.
29:05 Speaker 00 Big part of your job as a founder and a part of a job of the people you will be hiring is making sure that everybody uses the product, really understands it. Train, roll it out consciously. And this is different for every different industry for, you know, onboard them really thoughtfully. Maybe that's in the app, walking them through steps so they try different things. Maybe that's actually a person sitting next to them. I hope you caught this, but a very small kind of throwaway comment that Satya said earlier today is that one of the most like growing roles at startups is these forward deployed engineers, which I think is a really fancy term for just like boots on the ground people to sit next to your customer and make sure the product's actually working for them. Right? Whatever it takes.
29:45 Speaker 00 One thing I said a lot in my company, I still feel this is very true, is that your product isn't just the pixels on the screen. It's not just what happens when you click this button. It's the human interactions with your support, customer success with the founder. It's training, it's, you know, everything around it. If you don't get that right, then you might have the best pixels on the screen, but you'll be beat by a company that, that invests more in their customers and making sure that their products are actually well-used. That's all you need to do to build a fucking awesome AI app and beat our 650 million dollar figure handily. All right? So, open up for questions.
30:26 Speaker 02 Hello, thank you so much for your talk. I wanted to ask about the process of choosing what kind of industry to go into to try to create more automation in that way. So like, if there are already competitors in that space, would you suggest looking at another industry? Or would you suggest trying to dive deeper into a niche of that industry? Or like, what would you advise in that situation?
30:50 Speaker 00 So I don't think you should care about competitors at all. First of all, for some of these spaces, the market is so big, because we're talking about like, how much, how many trillions of dollars are being currently spent on like marketing professionals, or support professionals, or whatever. There's not going to be a single company that's going to win this entire market, for the vast majority of them. And frankly, a lot of the times, you're going to be, at first scared of your competitors, and then after you start building it, you're going to be dumbfounded about how bad they are. And you're going to outbuild them, you run circles around them, it's not about the competitors. But what I will say is like, kind of diving deeper into like, how to pick a market, the things I look at is, what are the kinds of roles that people are currently outsourcing, say, to another country? Right?
31:34 Speaker 00 If it's something that they're willing to do that for, then that's probably a pretty good target for what AI could take over. If it's a role where they feel like it's part of their identity to do it in-house, you know? For example, I don't think you're going to outsource for Pixar, creating the story of a Pixar movie. Right? That is, that is their, that is, they feel, whether they're right or wrong, maybe AI in two years will just like, do better Pixar than Pixar, but the people at Pixar are going to feel very strongly about the storytelling element. So, you know, don't try to outsource that part, try to, try to find the parts that are already outsourced. For example, find big markets, find where there's a pain point across many different companies. Find, find things you know about, or can get access to information about. These are the kinds of things I'd be looking at while trying to pick a market.
32:19 Speaker 00 But honestly, like, there's so many huge markets, you could literally just print out like, all the knowledge work stuff if you wanted to keep it digital, throw a dart. At everything you point out, whatever the dart lands, just choose that market and start running at it. And I think it probably hit a trillion dollar market. So competitors or not, don't care.
32:36 Speaker 02 Thank you.
32:37 Speaker 03 Perfect. Thanks a lot. So, Michael from So Strong. I have a quick question because you're a successful founder and many of us are going to found companies here. I wanted to know how has your focus changed across the different stages of companies? From say the pre-seed, what did you focus on versus, you know, the seed stages to the series A stages? And finally, to the exit and which part did you enjoy the most?
33:01 Speaker 00 It's a great question, Michael. So I'll answer what I should have done and also what I did do. All right.
33:05 Speaker 03 Perfect. Thank you.
33:06 Speaker 00 What I should have done is at the seed stage, focus on making a great product that gets product market fit. And then at the series A stage, focus on making a great product that gets product market fit. And then a series B, focus on making a great product that makes product market fit. And then series C, great product makes, you know, you can see probably the pattern here. What I ended up doing is I ended up focusing on all kinds of other things that didn't matter nearly as much as those things. And I think if you start from like, you know, because what is a company outside of its product? Like it's literally the service you're providing to your customers is through the product. And if you focus almost entirely around that and become obsessive around that, in my opinion, then a lot of other things will follow. For example, what people do we need to build a product that gets product market fit? Now you have like HR and recruiting, et cetera, to fill in for that, that answer.
33:51 Speaker 00 How are people going to find out about this amazing product? That's marketing and sales. What culture do we need at the business to create a product that people love and really use? Now you have, you know, other parts of HR and setting the culture, which is a very important part of your job as CEO. So you end up as CEO focusing on all these different aspects by necessity, but all to that one end. And what ends up happening for a lot of founders, because they read like medium posts and blog posts and they talk to their series A and series B investors, is they're focusing on HR or finance or fundraising or whatever, not as means to the end of grilling, creating a big, great product that gets product market fit. But instead, as an end to themselves, like, oh, we need to have a really great culture in the abstract.
34:33 Speaker 00 Or we need to like, now we need to hire marketing and sales. I did this. I fell into this trap. Big mistake. You know, I would, I would instead, and this is, I'm very, like I said, I'm one founder is very biased towards the product, etcetera side. But I think, I think personally.
34:47 Speaker 04 Hi Jake. So when I was 14, I sold my startup to Deloitte. And like you, I'm kind of looking for the next thing to do, like in the exit acquisition stage. If you were here at Y Combinator Startup School, what would you be doing tonight? You know, bar, case tax, whatever you're doing, what would you be doing here exactly tonight now that you exited?
35:05 Speaker 00 That's kind of amazing. I exited at 40 and you exited at 14.
35:08 Speaker 04 Yeah.
35:08 Speaker 00 So you're already well ahead. It's fucking awesome. Actually, I feel like in some ways for us in the early days focusing on legal made sense for us because I knew legal. But also it's kind of a mistake because at the time, the legal software industry, pre L L Ms is actually pretty small. Because it was like a fraction, like, you know, lawyers make a trillion dollars a year. It sounds pretty good. But how much of that are they really spending on software? And the answer is like a very small amount. So no matter how well we did as a company, we just weren't going to make something that really changed that many lives, that really made that much money ultimately from a business perspective. And we were only making incremental changes to the workflow and outputs of the people we were serving, pre L L Ms.
35:52 Speaker 00 And post L L Ms, all that changed, we were all of a sudden helped many more people and made them a lot more effective and changed many more lives. And I will tell you, having existed in both spheres of making small impacts on a small number of people and making only small differences in their lives, contrasted with, you know, making a huge impact on many more lawyers in our case, making them way more effective and efficient, replacing some of the work they were doing with L L Ms. The latter felt a lot better and I'm kind of addicted to that now. I'd be focusing, long story short, I'd be focusing on the biggest problem you could possibly think of that is possibly solvable with the technology and skillset that you have. You know, like what do people want? People want, what do businesses want?
36:36 Speaker 00 People want to be like skinnier and not like lose their hair. They don't want to do their laundry. They, you know, want to have, everybody wants to have a cleaner show up at their house for eight hours a day and clean their whole house and make it spotless. But you just can't afford to do that. But could you make a robot that does that for you? Right? Is that a kind of product that can serve everybody in the world? In fact, is that the kind of product that like the dishwasher in the 50s could unlock a lot of human potential? Because now people who are staying at home to try to take care of the kids are not having to clean up the house anymore. Right, because they can buy a thousand dollar a year robot or whatever. There is so much you can unlock with like, just thinking, what is the biggest problem that most people face? In businesses, you know, they want to market product, they want to sell their products. They want to make sure that people are doing great work.
37:19 Speaker 00 They want to replace certain parts of their work with like more consistent, more available. Like, that's where I'd be focusing my attention. It's just who's a huge problem that a lot of people have that you feel like you can solve and just go after it. Run as hard as you can.
37:29 Speaker 04 Great, thank you.
37:30 Speaker 00 I think I have time for one more.
37:31 Speaker 05 Hey, I'm Sibo, and I was wondering, if you're making AI to be an assistant or a replacement for a human, you could price that service based off how much time it saves a human or how much you would charge the human for as a salary. But if you're making something that AI is doing that humans could not possibly do, like looking through hundreds of thousands of like law documents per se, how do you price such a service?
37:50 Speaker 00 Yeah, I want to be like really nuanced with what I said earlier. I think at first you can start charging what the humans charging. And then you'll have competitors and they'll come in, they'll charge a little bit less. And then other competitors will come in, they'll charge a little less. And it's kind of beautiful, it's how capitalism works. And it'll make the service cheaper and cheaper and cheaper and cheaper. And at a certain point, you know, if unless you're in a very protected kind of space, you will be end up charging a less than the people were, which I think is probably a good thing at the end of the day for society. Right? Bad for your business, good for society, because now you can have the services of a lawyer, but for like 10 cents on the dollar, 1 cent on the dollar. For that new category of like, you know, I would, I would start from, what's the value? What's the value that you're providing to the business? Start there.
38:31 Speaker 00 If they're going to save a hundred million dollars doing this, or would have paid five million dollars to do this, okay, take 10% of that, 20% of that, you know, have a conversation with their customer, how much are you willing to pay? To solve this problem for you is probably the best place to start. I actually have time for one more question, rapid fire, so super fast one.
38:45 Speaker 06 Hi Jake, congrats on your exit. I know you probably get this question a lot, but when you're building things with prompts that are based off of models that may not be proprietary, how do you build defensibility and not end up as a GPT wrapper, basically?
38:58 Speaker 00 My fastest answer, just build it. And as soon as you build it, you'll see how fucking hard it was to build it. How many little pieces you have to build, how many data integrations, how many checks? How fine tuned the prompts need to be? How you have to pick your models super well? And when you do that, you're going to find that you built something that nobody else can build. You spent like two years just doing nothing but that, so I'm not scared. Don't be scared. All right?
39:21 Speaker 06 Thank you.
39:21 Speaker 00 Thank you, everybody.
Previous Transformers: The Discovery That Sparked the AI Revolution Next How To Recruit The Best Engineers And Sales Reps Page updated October 12, 2026 · transcript and summary are machine-generated
