Y Combinator CEO on How AI is Changing What Startups Build

0:05If you want to know how AI is going to change the workplace, one of the best places to look is at startups. And there's probably no one more steeped in the world of startups than this guy. Hi, my name is Garry Tan and I'm the president and CEO at Y Combinator. If you haven't heard of Y Combinator, you've definitely heard of some of the startups it's helped launch. Airbnb, DoorDash, Coinbase. YC, as it's commonly known, is a startup incubator. Gary says it gets about 100 ,000 applications every year, but only a few startups get accepted. Those that are get up to a half a million dollar investment.

0:46Gary prolifically posts about his AI use, and he says that AI is changing everything about how companies are being created these days. And it's also changing how he works as a CEO. Last month, I got on a call with Gary. He was in San Francisco, and I was in a studio overlooking Times Square. If you want to see a video of our conversation, head to Spotify. You use AI agents a lot. You've talked a lot about ways you're putting it into your life. What are you having it do, and what do you find that it's most useful at? I mean, I think of it as like self-driving rocket for the mind. Interesting. Like back in the day, you might have to have a team of 100 or 1 ,000 people to get done something that now only takes like 10 or 15 people.

1:33And so, you know, I think that incrementally this means that we need to boil the ocean. We need to increase the ambition of founders. And there are just more things that people should be doing. How long do you think it'll be until AI is as good at your job as you are? It can't prompt yet. So, you know, I'm on the lookout for that. I actually stopped drinking in February because I found that, you know, even having a single beer would make it much, much harder to think about code and like sort of run coding agents. Um, but you know, I, I would, I would say I'm going to start drinking again when, uh, the coding agents don't need me anymore.

2:22And, uh, I would guess that that's sometime in the next like six to 18 months. Wow. That's soon. Yeah. I mean, are you worried though about your own influence? I think the most important thing at that point will still be the most important thing. It'll be knowing what to build. It'll be like, you still need a human. You still need a human who knows other humans. you still need to understand context. You still need to figure out, well, what do people want and why? And then how do I build a thing? The mechanics of it get easier and easier, but knowing what to build and knowing what to want is still a very, very human thing.

3:03Welcome to The Journal, our show about money, business, and power. I'm Ryan Knudsen. Coming up on the show, a conversation with Gary Tan about AI and startups.

3:25This episode of The Journal is brought to you by Harvey, an AI platform designed for legal and professional services. Built and tested by lawyers, Harvey is trusted by more than 60 % of the AMLA 100. The platform dramatically reduces time spent on research, drafting, and document review without sacrificing quality, all while meeting the highest industry standards for security and compliance. Harvey, AI tailored for law. Visit harvey.ai to learn more and request a demo. This episode is brought to you by Claude Cowork. Cowork brings the agentic power of Claude code to your desktop so anyone can use it and have Claude do real work.

4:08Point it at a folder or connect tools like Google Drive. Describe what you need and it handles the rest. Organizing files, building spreadsheets, drafting reports from scattered notes. Even use it to plan out your week and stay ahead. Hand off the tedious work. Get started at claude.ai slash the journal.

4:33So I want to talk mostly about AI today. There seem to be two extreme views of where AI might take us. On the one hand, there's people that worry about an AI apocalypse, that it might bring about the end of humanity. And there is, on the other end of the spectrum, there are people who think that it could bring about a utopia in the future. Where do you fall on that spectrum? I mean, I think it's going to be different. And Utopia is probably too – like things are going to change. And anytime the cheese is moved, there are going to be winners and losers. And then net-net, my hope is that there are more winners than losers because it's just a really, really powerful technology.

5:15And from our perspective, you know, we're trying to help people start new businesses. And I, you know, deeply believe that doing that right means that you're solving problems for other people. And so in aggregate, like, you know, this kind of technology in particular makes that way, way more possible. And so will people lose jobs? Yes. On the flip side, though, with AI, it's like you used to have to hire a lot of people and raise a crazy amount of money. And now you're actually able to create businesses for far, far less. And there can be a lot more of them. So in aggregate, I think humans can solve more problems.

6:01And there's a lot more entrepreneurship that can happen. Like our standard of living will rise. and things will just actually be the way they should be. And maybe that is utopian. So obviously Y Combinator, you invest in startups, you nurture startups. In what ways are you seeing AI allow people to start startups more quickly? I mean, the most simple version of this is you're able to write, like one person is able to write a large multiple of what they used to be able to do. So, you know, I run YC today. About 13 years ago, I was a designer in residence, and I was building software for YC. And in those days, I was writing about 15 to 20 lines of code a day.

6:48And, you know, that's not a lot. But, you know, now with agents, I've been doing something closer to 400 times that. And so you're able to – one person is actually literally able to do the work of 10 to 100 people previously. And so what that means, we're sort of seeing dozens of companies in the last few years go from zero to tens of millions of dollars in revenue with like less than 20 people having only raised a seed round of like under$5 or$10 million. So a lot of this is like unprecedented. Like you've never seen people be able to do that before. And this is great. Like, this is all great news for people who want to start companies.

7:29It's great for people who want to start companies. Maybe not so great for people who want to work at those companies. But how much— But those people should start companies. How much do you think that that can scale? As a business gets bigger and bigger and bigger, at what point does it need to start hiring more people? Or will it ever? I mean, this is where I think this requires a change in mindset, right? if you just want to maintain things as they are, you just want to sort of let people go. But I think that the world is actually so full of problems that we're entering this other age. And this might be a change in the nature of types of companies that can be started.

8:14Gary says that the types of companies that Y Combinator had traditionally funded were software companies. But thanks to AI, YC is now working more in hard tech, Things like nuclear reactors and defense. We have a company called Toranol that kills mosquitoes using drones, for instance. Did I see a video of them recently where there's a little tiny drone that can... Oh, yeah. What does it do? I don't know how... Yeah, it's got computer vision. And it just zaps the mosquito? Oh, it just flies right through them. And just kind of cuts them up. Yeah. Huh. And so you have an array of them in your backyard and, you know, you just don't have mosquitoes anymore.

8:57That would be nice. This is a very concrete version of this. It's like five or ten years ago, this is a company that maybe wouldn't have raised money. Like it just would have been too expensive. But what if someone wants to make a company and they think of us as the mosquitoes? I mean, I'm being glib. Well, then it'd be illegal. Right. Then it'd be illegal. So, you know, I'd be the first to say, like, if this is starting to, you know, destroy society or, you know, be deleterious, like, hey, maybe this is something we should slow down. But I just see too many places where this is actually so useful and so powerful and so empowering.

9:32You know, here's a startup that if the cost of doing it became viable, if you can do it like better, faster, cheaper, and then it becomes something that is ubiquitous. Like, you know, I would love to be able to go outside and not be covered in bug bites. Are there certain kinds of companies, people that have startup ideas that you say, we're not going to fund that or work with that because we think that that idea of yours, while interesting, is maybe too vulnerable to AI? honestly everything is so brand new like the rule book has been thrown out like three or four years ago um you could argue that there were sort of things that um you know thousands of founders have gone after and tried to do and you could just safely say no that's not going to work like a thousand people have tried that but right now um you know i think we're getting into a moment now where it's like actually all the rules have been rewritten and everything is new again what about your company, Y Combinator, how much are you relying on AI to make decisions that a human used to make?

10:38Every single person at YC now has their own personal AI agent. And, you know, we have a company brain that actually flattens access to all the information so that, you know, it's actually super democratized. Have you run into any problems with that? I mean, sometimes agents can, you know, run amok a little bit and do things that you didn't intend for them to do. Oh, yeah. I mean, the thing is that you can just have watchers on that, right? I mean, there's actually a huge cottage industry now of like, how do you make software that, you know, is watched by other AI? And so, you know, we funded a company called Clawvisor, for instance, that is sort of a watcher of what your open claw does.

11:22And so if you want to access Gmail or make a draft, it itself is an AI that is watching that AI. And saying, hey, and if it starts to stray from its objectives in any way. Yeah, then it sends you a message. It's like, hey, looks like it was trying to do this. Is this what you intended or not? Interesting. I mean, you always hear about these stories about, oh, the agent went and deleted my whole hard drive and things like that. I mean, we tend to be very careful about right access. And then when I'm actually sending a text, when I'm sending email, when I'm sending Slack, people often joke these days.

11:59It's like, oh, Gary, like, is that you? It's like, yeah, yeah, it's all me. Like, if I send you an email, like, it's me. If I send a Slack, it's me. You never let your agent actually press send on any message. That's right. Yeah, I have to, I copy and paste and I send it myself. I feel like that, you know, that's my covenant with people out there. After the break, more with Gary Tan on the less utopian side of AI.

12:45and everything else. So healthcare is connected, not complicated. That means cheaper prescriptions that are easier to get and care that looks at the whole person, how you need it. Optum is making healthcare work as one for everyone. Learn more at business.optum.com.

13:06So there's a growing backlash to AI in some corners. You know, you're seeing a lot of opposition to data centers. You're seeing even executives of AI companies face threats of physical violence. How do you talk about that issue with your startup founders? And what do you make of that? I mean, we certainly have a responsibility to, you know, the people when we build these companies, right? Like, are we taking care of people's data? Are we, like, trying to be benevolent? Are we trying to create, like, something that actually solves legitimate problems? And if you do that, that's great, you know?

13:46I mean, I think AI is, like, AI is the most powerful new technology, obviously. And it can do a lot of great and a lot of good in the world. A lot of people don't think—I mean, a lot of people worry that it won't. Right. I mean, how do you bring more people on board and, you know, how do you try to solve this issue where you're getting people who, you know, are dedicating their lives in some cases to trying to stop the development of this technology. I mean, I wish they would just use the technology themselves, right? Like there's new innovation that is happening, whether it's math, physics. I mean, they're just, I mean, that's powerful.

14:28And that's pushing forward what humans are actually capable of doing. And what I hope is that that translates into like real medical breakthroughs, for instance, like being able to solve cancer, being able to actually make way on clean energy. Do you think it's going to take a breakthrough like that before more people get excited? I mean, it's happening. It will happen. And I think the fear is not totally unfounded. I mean, even the most accelerationist people I know are worried about, you know, can Skynet happen? But, you know, I love movies from the 90s and 2000s as well. But that's not the only world that we could possibly live in.

15:09I mean, that's the premise of the Terminator, obviously, is go back in, you know, Terminator 2, go back in time, try to stop this company from making this technology. Does that keep you up at night at all? I mean, that's a movie. The aspect of violence? I mean, that's a movie and it's sci-fi. Well, sure, that's a movie. But I mean, there are people, you know, someone threw a Molotov cocktail at Sam Altman's house. I mean, like, does that concern you on a personal level? Absolutely. I mean, I think that, you know, both political and I mean, it's basically another form of political violence. We already have laws that constrain the things that are deleterious in society.

15:49I'm not saying that we don't need new laws eventually, but it's so premature. It's so early. And right now, like, we're in the steep part of the curve. To scaremonger and scare people away from the super powerful technology is just, it's sad. Like, instead of, you know, instead of trying to stop it, like, use it. And so if we took, like, even half of the energy that's going into sort of scaremongering about AI and actually just using it, like, let's solve these problems. Like, there are just too many out there to solve. And that's the world I want to live in. Gary is very active in local politics in San Francisco and California.

16:28One of the things he cares most about when it comes to regulating AI is making sure everyone has access to the best models, which right now are controlled by OpenAI, Anthropic, and Google. One of the things that we've been tracking is sort of this fight between big tech and little tech. and this always happens. You know, you could look at an Apple or even a Google and say, well, as time goes on, they want to capture more and more revenue. So they will actually close their platforms to others. And that's probably the worst thing that can happen to tech and to tech innovation. And so that's probably the number one thing that I'm most focused on.

17:09I mean, obviously, Y Combinator is unique in that we're sort of like the defining union or guild for early stage startups. And the thing that allows them to succeed or kills them ultimately is access to markets, platforms, APIs, and ultimately LLMs. And so we really, really care about access to all of those things. Access meaning that you don't want OpenAI or Anthropic or any other company to put up some rule that says you can't use our technology for X, Y, and Z reason or that it should be a certain price. Yeah, users should have choice. And that choice actually happens through lots and lots of small startups, all offering better and better things that are in competition with one another.

17:53That's how consumers get choice, and that's how technology should work. And it certainly isn't how it works yet. So are you saying that you think that like the big AI companies now, should they have like open source versions of their models? Because now, I mean, the way things are going, it seems like this will entrench the big tech companies at the center of the economy and give them an enormous amount of power. If there's suddenly all new businesses are being built off of their LLMs, how do you avoid that future? I mean, ultimately, I think there's a debate happening right now in the halls of power in D.C.

18:28about whether or not people should be allowed to have access to Chinese open-weight models. And I would argue very much in favor of that people should have access to that. Like that is probably the biggest democratizing factor for no lock-in. Because it gives more competition to American companies or what's the argument? It just gives competition, period. It gives choice. And so I guess there's two things. One is that the LLM itself is turning into a little bit of a commodity. Like, we'll sort of switch to whatever the best one is, and we sort of hope that that remains an American company. And then there's a lot of tension right now around, you know, as we talk right now, like, Kimmy just came up with a Chinese model just came out.

19:13And they're about to release open weights on something that is almost as good as the very best American models. Would you switch to a Chinese model? With the right guardrails, I would. Because you can actually look at, you can instrument everything about what goes in and out of those models. And, you know, once again, like, all AI can be safe enough if you watch it and if you put it in the right harness and watch it the right way. Why does it matter then if those are American companies? I think that intelligence is so important that it's really about access. If we want American companies, we want American innovators, and we want American defense companies, for instance, to be able to use whatever the absolute best is.

20:01And so, you know, when you're ultimately talking about, I mean, you're talking about dominance at the sort of societal level, right? Like whose companies are the most innovative? Whose products are the best? Whose military is the best? And I think that over the next course of decades, all of those will ultimately be answered by who has the most intelligent models. For the past few years, the headlines on AI have been all about the inevitability of mass layoffs. It's something we've covered a lot here on The Journal, and we hear about it from listeners a ton. But it turns out that the transition to AI at the corporate level is happening more slowly than expected.

20:42And Gary thinks that's a good thing. Society will absorb this. I think there will be transformation in every organization, but it'll happen slower than we might think, and that might be a good thing. Like when you look at a Fortune 500 company and you think, oh, it's going to take forever. Like that's not a bug. It might actually be a societal feature, right? The organizational latency, just sort of the inertia of big companies, it may well be the saving grace for society at large. I interviewed somebody recently who compared AI to being a conductor. and he was like, you know, he worked in marketing and he was saying how in marketing, you know, you used to just think about, okay, I want to play the drums and that's, you know, email communications or I want to work in strategy and that's, you know, playing the flute.

21:34And now everybody needs to think of themselves more as conductors because they can use AI to do all that. And I thought that was a really interesting analogy, but it also made me think that, you know, not everybody wants to be a conductor. Some people do just want to like wake up, put on their, you know, put on their shoes and go hit the triangle. And so like, what is that going to do to society if people have to, if in order to succeed, you have to level up in terms of how you even think about what a job is. And a lot of people may not want to do that. I mean, it does, it seems like we need to be more creative about what these jobs entail and look like.

22:11And I mean, on the flip side, like maybe that's where live events come into play. Like it's possible that we should be throwing a thousand times more parties. Yeah. Like, I think that what are people going to do with all their time, right? Like they can actually focus on the arts. Like, I mean, I always think about, you know, Aldous Huxley talked about in the 20th century that, you know, why are we doing 40 hour work weeks? Like maybe it should be 20. Like these things become more and more possible as the standard of living can rise. All technological innovations up until this point have not really resulted in less time working though do why do you think that this one might right it's um i mean what's funny is i sort of know for a fact that most people who become very ai pilled uh myself included like work about two times more yeah i feel like i've heard that from a lot of people once they once they realize what they can do they like can't stop they're like oh my god i can do this and i can do that and if i built and they just like it starts to envelop their life as they start to do more and more with it.

23:14But that's obviously, you have to have that sort of drive inside yourself, you know. That's right. I mean, it might just end up, you might just end up having people who are like, very, very productive and other people who don't have to be, right? Like, my hope is that it unleashes a brand new, just a firestorm of entrepreneurship of like, there are so many problems out there that we could solve for one another. And then maybe they don't have to be all billion dollar companies. They might be actually a really nice bar where you have community. Maybe it's a return to third spaces. I mean, there are just so many things that people like and enjoy.

23:58Going back to YC, right? Is YC going to be run by one person and a whole bunch of AI agents. Like the reality is like, no, never. Like we want to bring together really smart founders who have been there with the next generation of people to do that. And so the people are the point. And if I expand that out into broader society, it's like we should have better restaurants. We should have better social services. We should have better cities. We should have more housing. Like what are the things that people actually want? Let's work backwards from that. and more and more people can go towards that.

24:36We will never truly achieve utopia, but that doesn't mean that we can't try. Well, thanks so much for your time, Gary. I really appreciate it. Thanks so much for having me. Great questions. That's all for today. The Journal is a co-production of Spotify and The Wall Street Journal. If you like our show, follow us on Spotify or wherever you get your podcasts. We're out every weekday afternoon.

25:01thanks for listening see you tomorrow