#2281 Garry Tan: Y Combinator Startups Growing 5X Faster – Here’s What Changed

0:00Hey there, Freedom Fighters. My name is Andrew Warner and this is a new series for me. It's called The Next New Thing. Here's what's up in this interview, then an intro, then we'll get right to it. How do you compete with these bigger players? We're seeing routinely YC companies with 10 or 20 people get to 10 or 20 million dollars a year in revenue in 10 or 20 months. That's like literally never happened before in software. Talk to me about how you use AI in your video creation. I took the scripts of all of the top videos that I ever made from my YouTube channel. I'd throw it at this prompt and then it would generate these beautiful three-act narratives.

0:38I could have a new 10-minute script ready, whereas it normally would take me like several hours. How are you changing Y Combinator? Let's stop competing with all the other VCs. Let's be their partners. I'm going to ask you to do something you're uncomfortable with. Oh yeah, what's up? Gary Tan is the president and CEO of Y Combinator. The next new thing. Why don't we start with the case text story? Because I feel like there's a before AI for that story in AI experimentation. And then once it took AI, everything changed. I worked with Jay Keller, the founder of case text back in 2012, 2013, when he first went through Y Combinator.

1:16And that was also my first stint at YC as a partner. And they were sort of doing basically your web 2.0 for law. So literally what's happening with case law and new legislative, I mean, whether it's legislation or literally judgments, like all of the documents that the legal profession throws off, they would index, which would help you understand the law. And that was really what they built for something like going on 10 years. It grew by SEO. and Jake's both a great technologist and a great lawyer. And so he was really able to go into that market and make something based on what was happening in society and in tech at that time.

2:02I think there was also like a Q &A component of this, right? So they could go and talk to other lawyers. We're going to get into how things get better. Why wasn't that enough? Some things can become huge and drive billions or tens of billions of dollars in revenue every year. and some things really could only get to, they only provide value that, you know, and then you multiply it out by all the people who need it. And that might only total up to 10 or 20 or 50 million. Like that's weirdly quite common. I think a lot of founders are worried about that early, but my sense is maybe it's premature worry because embedded in that is also the case text pivot that, But they got users and an understanding and a useful corpus of data, all of which turned into a tremendous moat for them literally right at the correct moment.

2:52As technology itself shifted, something that could only make tens of millions a year could suddenly become something that could make hundreds to billions of dollars per year. And that was the dawn of the large language model in 2023. Tell me that story of how they came up with that. The cool thing about YC was that Jake basically had access to early versions of ChatGPT, GPT-3. These were sort of toy earlier versions of it. And they were certainly astonishing and interesting. But they were not useful yet because the LLMs actually would just hallucinate. They were early in the journey, so there wasn't enough data.

3:34The number of parameters, the sort of size of the models was too small. and that's what Jake found as he tried to use large language models to do a lot of the things that you and I take for granted today. Right at the dawn of this stuff, it was not that useful. You know, it was sort of a horseless carriage, if you will. It was an oddity. You could look at it and say, well, maybe this will work, but it's mostly a toy and nobody will actually use it yet. possibly ever, right? And certainly when I first saw it, I'm embarrassed to say like, even as an investor and technologist myself, it's like, that was the consensus at the time.

4:15And at that moment, at the dawn of large language models, that was correct. Like you couldn't use it for useful things yet. Jake being a great designer, engineer and lawyer, he tried really hard to make at work. And it would hallucinate. And, you know, he also was operating in an area that, in particular, really had high sensitivity to hallucination, you get one thing wrong, and you're fired as a lawyer. So, you know, his his particular space was fascinating to me, because it particularly could not withstand any hallucination. And as technology curves and cost curves go. This was something that I think surprised everyone.

4:55If you were Greg Brockman or Dario Amadei at that moment, you started internally, you were talking about the scaling laws and that the loss function was going down as log linear to the amount of data and compute you were putting in. And that was an astonishing realization that like there was a path to potentially AGI or ASI. The rest of us on the outside had no idea. And I think Jake also didn't have any idea. But because he was in the YC community, OpenAI itself was a spin out from YC research by Sam Altman. So then he starts adding this on once it's ready, once it's ready for lawyers. What was the original use?

5:39And then how did it take off? I believe he basically started using it for being able to answer specific questions about legal cases. And once he got access to GPT-4, he realized that if you cut down the size of the question to small enough, and today we call that context engineering, but at that moment, he realized if you asked a very long-ranging question, like, is the defendant guilty? It's such a big question that even GPT-4, I mean, you could argue that some of the reasoning models today are actually much more capable of doing it. But back then, you didn't have multi-stage test-time compute reasoning.

6:26at that moment, if you chopped it down to a bite-sized chunk, like you gave it some amount of context that a human being, given the same context and the same prompt, would answer in a certain way, he found that he could, given inputs and outputs, have output that was usable, useful, and reliable, and not a hallucination. But it required you to chop that down into a particular small enough step. I think of Jake a little bit like the first man on the moon. You're like, oh, you can chop it down. And then you should actually have tests for a bunch of different inputs and outputs. And you should have evals that actually give you a sense and certainty about specific tasks.

7:12So you would sort of do tailored time and motion study of exactly how a lawyer might, you know, a lawyer often has to do a timeline, for instance. So what he would do is like chop it down into what would I do as a human being? Well, I would start skimming each chunk, whether it's a sentence or a paragraph. And then I would try to score it based on whether it was noteworthy or not for do I need to put it in a timeline or not. And so you can imagine, I mean, he's like sort of very logically creating some of the first ways to do prompt engineering. So he's taking this technology and he's finding a way to make it useful by thinking almost like a human being, a little bit like a machine, and then coming up with the answer.

7:58He now has a better tool. That gives me two questions. What happened to the business? Did it immediately go from like sales cycles of, I think I saw you say in a video, a year or more to now a month or less? I guess the really interesting thing was he built the first versions. It could do like, let's say it could do a, it could come through thousands of pages of documents and give you an accurate timeline of events, for instance. And that's something you would hire a legal analyst or associate to do. And it would cost thousands, tens of thousands of dollars. Right. So instantly you can say, I will save you this much money direct to the bottom line.

8:34Do you want to buy? And it becomes an easy answer. I think that was the feat of strength. Like they took the corpus of Enron emails, for instance. And I think the example Jake likes to use is you could ask questions about emails where it involved like high nuance things. Like you could ask it about ironic jokes that the CEO had made. And it would be able to like discern like, oh, they made a joke about their fraud in this particular way. And it would like find the reference for you. So that was sort of a demo that they would show to lawyers and it would just be so astonishing that people would say, I need to buy it right now.

9:15This is the future. And it turned out to be very – like not just a little bit correct. I think I'm still astonished day to day, especially as new model releases come out. That explains why every time I see what it does, I will see – and it detects sarcasm. And now I get why that comes up. Then the other thing that comes up for me, Gary, is we keep hearing that maybe it's useless to create anything in this space because the big companies are just going to take it on, right? You can imagine a world in which ChatGPT does all this or Google's Gemini and its consumer-grade products that people are using to figure out what to do with their weekend plans that it's natural for them to then ask the same question for.

9:54It's not an unfamiliar tool. When you do this, how do you compete with these bigger players? Yeah, absolutely. I mean, I guess I have two answers. One is obviously the one that we're, you know, at Y Combinator, we, you know, we believe in that we see it happen. Like we just see small teams of people go out into the world and create, you know, they're obviously using these incredible frontier models, but they're adapting them to the very specific things that real people in the economy actually need. And so they're not necessarily glamorous scenarios. They're customer support scenarios for HVAC consultants, for instance, this fragmented industry, but a very big industry.

10:42And they're building software. There's a company called Avoca that we work with at YC. They're doing exactly this customer support for HVAC. But V1 of it was basically Service Titan. So Service Titan is an incredible public company, but they're basically software. And HVAC consultants and firms spend about 1 % of their dollar wallet. For every$100 they bring in in revenue, they spend about$1 on software like Service Titan. But they spend$5 or$6 on actual people picking up the phone and doing scheduling and doing all that stuff. So the wild thing that we're seeing is that if you like scope what you're doing and make the thing that is perfect for that set of people, you can't just take ChatGPT and have it do this type of work yet.

11:37I mean, it's entirely conceivable eventually, but it hasn't happened yet. And while that is still true, people are sort of building the next service titans. And then the wild thing about it is Service Titan is an incredible business, but then you could have something that takes over that 1 % and then expands like 5x or 6x bigger. So that's sort of why we're seeing routinely YC companies with 10 or 20 people get to$10 or$20 million a year in revenue in 10 or 20 months. And that's like literally never happened before in software. You know, the average rate at which YC companies grow their revenue during the YC batch, which is a 12-week process, is 10 % per week on average.

12:25I mean, some of the batches have been growing 15%, 20 % a week. But it's been at least 10 % a week for more than a year. And so there's something in the water. Like, there's something happening. What was it before? I thought it was 10 % a week through going. Oh, no, you'd have one or two companies grow 10 % a week. That was like the aspirational, like, if you could do it, that would be what good looks like. And then now on average, everyone does that. So it went from on average 1 % to 2 % to 10 % to 20 %? Yeah, I think the average was probably close to 2 % to 4%. And then now it's consistently 10 % to 20%.

13:03And this is in revenue? Yep, in revenue. So, and, you know, it all goes back to the case tech story, right? Like before, it's like, yeah, I know I need to replace my software. You know, oh, I'm still using SAP or I'm using whatever I was using. I'm still using spreadsheets. Like it's, you know, having better software was much more of a nice to have. Like it's something that you felt like you needed to do. And then today it's becoming, oh, like I see a demo. It's really impressive. I could see how that would hit my bottom line or basically create a better product or service immediately. and then, yeah, I need it right now.

13:42When can you start, right? I totally feel that. I feel it in the air. Then it kind of brings me back to the conversation that I had recently with the founder of Read.ai. This is David Shim, and I said, when I do sales, I want a note-taking app that keeps guiding me towards closing a sale or at least analyzes me afterwards based on a sale. That's a great scenario. Great example. And he goes, Andrew, that's not the way it's going to work in the AI world. What you're going to have is one tool, one note taker. He'd like it to be obviously read AI that does everything. And if you say I'm a salesperson, it'll customize some of the feedback that you get for sales.

14:20To me, if I even have to customize it, it's an extra step. And so I've been seeing this, Gary, in conversations with builders. Some are saying AI doesn't need to be customized at all. And I see you squint as I say that. So I think that maybe you have a strong opinion here. And others are saying, absolutely, what you are is go down to the level of HVAC. Yeah, I think that we are sort of, we might be at a moment where it's too early to tell. Obviously, the stakes are very high. But if we got to a point where AI is truly, you know, not just AGI, but ASI, it can like far exceed that of what humans can do.

14:59All bets are off at that point, right? So I don't know. I feel like this is almost like the reverse Pascal's wager for AI a little bit. It's like, well, it's entirely possible that ASI happens and what some people think will happen happens. But the society that we will need to live in will be reconfigured in such a radical way that will there be jobs? And at that point, the hope is that if we have actually access to clean and clean solar and wind and maybe even fusion with Helion and things like that over a 30, 50 year time frame, like society can reconfigure into one that's really focused on abundance.

15:50When you're talking about startups and competition and markets, like we still live in a market economy that is driven by, you know, should I buy X or Y? And, you know, I think it's like, you know, certainly in some sense a flawed system. On the other hand, it's certainly the best system that we have, like the invisible hand. I want to know which direction you think, but I'm looking at the list of companies that you at YC have helped launch just in the fall of 2025. What I'm seeing here is there is a focus. It is companies like, where is it, Market Silver Bullet for Trade Compliance, to give you an example of what I see.

16:28I see another one, Bluma, automating short-form video ads at scale. So you really are still saying, I'm going to be focused narrowly on a vertical. Am I right? Or am I just looking at a handful and drawing these? This is also about making individual founders successful, right? I guess famously, I think at some point, Sam Altman came out and was, while working on OpenAI, he was sort of rethinking whether the classic YC advice was correct. I mean, obviously, we're friends and we had some exchanges about it. I think that he's sort of changed his tune a little bit in that he's seen now that AI, like all the startups out there using his APIs are sort of his commercialization arm.

17:13And that's not a bad thing, right? There was a time when I think he said he just wasn't sure if all of the advice around make something people want and like being lean was quite the right thing. And then to me, I think looped had to be lean. You know, a lot of people who start really huge companies had to start companies that were much more specific. Elon Musk had to start Zip2. I think the reframe for us at YC is that we actually want people to be directly in control of their own destiny to the extent they can. Can we do that? When you're starting a company. The thing that was exciting for me is like I'm holding here.

17:59This is well told. It's a mug by someone who I don't think 20 years ago could have created a mug. It's beautiful. It's got like the city that I'm in. He sent me a bunch of them, Gary. One with every city that I've done Mixergy in, which has been a lot. Cool. Um, and I think about him a lot because that kind of entrepreneur couldn't have existed before, but now they do. You are a video guy, beautiful videos. You always had good taste in video. I don't think you would have existed, let's say 20 years ago, because it would have been forever to videotape, to edit, to put your spin on it. Do you think the same thing now is going to happen with software that more people are going to be able to create it and it's going to be sustainable or it's going to give them enough money to sustain their lives?

18:41Absolutely. I mean, that's certainly my hope. The reverse is like too dark to, you know, if anything like that's some of the reason why we spend a lot more time in DC. But Gary, even if it's the same, like I'm wondering that because of all the vibe coding apps, I keep seeing vibe coded apps from people. Will they turn into something significant or is it going to be like most of the YouTube videos where there's no business from it? It's just fun to create. Or does that even matter? My argument would be, I mean, especially vibe coding. The Claude Code team apparently writes 95 % of their code is written by Claude, which means very directly that each engineer working on Claude Code themselves is doing the work of 20 people.

19:24That's sort of a direct quote from a recent Lenny podcast with one of the co-founders. And so I think that that's actually the good news. You know, I think if you look at tech across like 10, 20, 30 years, it's actually that like the access to good software is incredibly inaccessible. And one argument I often make is if you use an iPhone, you probably have hit bugs in Apple Calendar. And it's like very frustrating because, come on, guys, like this is the built in thing to Apple, the iPhone. Like the iPhone is, the Apple is like one of the most dominant tech companies in the world. And yet they cannot find good enough software engineers to fix the basic bugs that still exist in Apple Calendar.

20:08And so that's been true for time immemorial. Like if that's true for Apple, how could you possibly imagine an HVAC person ever getting access to good software? And, you know, that's the difference today. It's like, hey, you can have it now and it can be customized to you. And if anything, the funniest thing is if AI and Cogen gets even better than it is today, that might be one of the vectors by which HVAC people compete against each other. You might even choose the one that has the best status and the best app that can tell you exactly when things are done. or if someone uses, they use Replit to create software or maybe there's a vertical version of Replit just for workflow for managing your workers, right?

20:58Like anything that you can imagine, it could actually create a better product or service. And then net-net, what this might mean is that just like everything that we get in our day-to-day lives is just better, faster, cheaper. And then more is more actually. Like, it's actually good that, I mean, sometimes to link the abstract to the specific, I'm like, a good example of this would be, I would love for every apartment in San Francisco, for instance, to have dishwashers and washing machines, right? like in a weird butterfly effect sort of way. Like if you think about people doing better work, more meaningful work, doing it on time and at the right time for a better price, like that sort of, that's how the market creates like higher, basically higher standards of living, right?

21:54And so that's sort of like what I hope is and what I think will happen is that as long as people can start businesses, they can make better choices, they can make better products. This is actually a much bigger engine for making our day-to-day better. You're imagining a world where instead of having HubSpot and Salesforce and Close and a couple of others that are really big, HubSpot, Salesforce, then you got them in, you're imagining a world where there is a CRM for podcasters like me, a CRM for HVAC, in fact, multiple of them. And so the revenue of Salesforce, I imagine, would then be spread across a bunch of companies, and there would be a bunch of companies whose founders are not as wealthy, but they are making a strong enough living.

22:43That's the world you see. That's right. And then on the flip side, even if they, well, I realize that's also often not how it works out. I also wonder if that's great for you because what you're looking for at YC is not to have a bunch of small companies where the founder can live a good life, maybe buy a second home, but where they're building the Mark Benioff size successes, right? Yeah, that's right. I mean, I guess YC is funny because even if someone doesn't end up making the Salesforce size thing, like they often sell their bit. I mean, that was true for Posterous. My YC startup ended up selling to Twitter.

23:23Our YC batchmate, BackType, it was Chris Golda and Mike Montano's company. They sold to Twitter and then that team ended up creating Twitter ads. I think like our old teammates at Posterous ended up making the Twitter, you know, making the first Twitter mobile apps, and working on that team. So, I don't know, there is like a creative destruction aspect, and then on a sort of day-to-day career basis, like it's better for people to become founders, learn how to create things for other people, and then either you manage to get product market fit, and you figure out a moat so that you can be as big a company as possible.

24:06or even if you don't, like everything about your life and career moves ahead by five or 10 years faster than it would have been. Or, you know, we have lots of friends who, instead of starting companies, they stayed at Microsoft. And it's better to be directly in the face of real users and shipping real code and product and then learning how to support that because that's just actually valuable. And I think there are lots of other jobs out there that are slow moving, bureaucratic. I actually sort of wonder, like, I was hanging out with another investor who sits on a lot of boards. And we were thinking, like, man, there's increasingly, like, a lost generation of people who work in big tech.

24:54And I think, well, even at YC, like, we are seeing the rate of 18 to 22-year-olds at YC is up by more than 100 % year on year. Because? The rate of 22-year-olds to 25-year-olds is up about 20%. And then the rate of 25 to 30-year-olds is actually down by like 10 or 20%. Why? And a lot of those people started their careers actually during ZERP. so they're actually sort of you know hanging on for dear life at um both startups and fang and uh those are also like funny enough some of the people who are the biggest ai deniers like they just don't believe that um the sort of revolution is happening you're saying they got such a good job where they were they don't want to lose it and go and kick off and start something different got it whereas the the young generation like they're incredibly hungry right now because fang is not hiring like the the zerp jobs are not there and a lot of them i think are are playing more with this it used to be the playing startup was fun then it became playing youtuber was fun but playing startups felt like now you're becoming the man but now you see lovable and cursor and all these tools and you say okay they're making it more accessible let's play with it and then you you lean and create something i told you earlier that i would give you a great example of why Y Combinator is like inner access.

26:19David Roganmoiser, he is a guy who started out with proof, a thing that was gonna help e-commerce companies. And I love after he got into Y Combinator, he goes, what we're gonna do is we're gonna revolutionize e-commerce. We're gonna make every Shopify store customized based on who the person is because we're gonna have a tool that goes across all of them. I go, boy, Y Combinator really gets people to think beyond the tool. Before he had a little frickin' widget, now he has this, changing the whole thing. And then suddenly, he created what became Jasper, and it's because he got to see the original open AI tools and he said, okay, I got to pivot everything.

26:53We're going to create an ad creation tool that uses AI. And then he changed it to based on user needs to create a writing tool for everything. That is the inside Y Combinator access, right? Absolutely. How do you stay in touch with people to keep guiding them after the time that they're in the program? Oh, well, I mean, these days, everyone who gets into YC, they have one particular primary partner. And obviously, when you apply, it goes into sort of this giant pool. But then we have 15 equal partners on the investing side who like, are in we basically are in there trying to fish, like we're, you know, in there with our nets.

27:32Like, let's take a look, like, let's watch the video. Let's try the demo. Let's read everything about the founders, what they've done and what do they know about their users, about the product. And then we try to figure out, well, who do we meet? And then anyone you choose, you meet. And then when you meet, it's up to you whether or not you accept them. And then when they're in, like, you always have at least one person who is sort of like your investor at YC. And that is for the life of the company. You're still getting on calls with them a year, two years later. That's right. So, you know, I basically, I think the bad version of thinking about YC is like, oh, it's like a summer camp and you have a camp counselor and you never talk to them again.

28:16And then the good version of it is like, oh, yeah, YC partners are, you know, actually renamed their title. It's like not group partners anymore. It's actually general partner. And we're actually investors invested in the future of the company for the life of the company, right? It's like having your best angel investor who is there for you all the time. And then that's for many years. So basically, on our end, the best thing we can do is understand the business, understand where the founders are coming from, who their customers are, and then just ask questions. It's like, could this be bigger?

28:50What's the most interesting thing that you've learned about your customer? What are things that get people promoted? What are things that are resonating? Let's double down on those things. Let's be super frank about what we've tried that doesn't work. Like if we're spending money or resources or, you know, people time or your time on it, maybe we, you know, we have that or set that to zero and anything that's working, like let's double down on that. And then you're just having someone who's outside of your day-to-day who could be a sounding board. I mean, there's that. And then, honestly, the batch itself is perfectly designed to help you have not just your partner, but dozens of other people who are all, like, the outcome of a top 1 % process.