Sam Altman: "Never a Better Time to Do a Startup"
OpenAI’s Sam Altman closes Startup School 2026 on ambition, agents, and the biggest window for founders in tech history.
In 2005, Sam Altman was a Stanford sophomore in YC’s first batch, building a startup in a little Cambridge office while Paul Graham cooked the founders dinner.
Twenty years later, as co-founder & CEO of OpenAI, he closed Startup School 2026 in conversation with YC's Garry Tan — on agents, ambition, and why the ceiling for what a startup can take on has never been higher.
Timestamps
00:07 — From YC’s First Batch to Today
03:04 — What Paul Graham Taught Sam Altman
04:17 — Why Startups Matter More Than Ever
06:57 — The Coming Golden Age of Ambitious Startups
09:30 — Why Startups Win During Technology Shifts
11:46 — Building OpenAI When Nobody Believed in AGI
14:45 — Finding People Who Share Your Conviction
18:05 — Help People Before You Know Why
19:43 — Earnestness, Ambition, and Ignoring the Haters
24:04 — The AI Safety Incident That Changed the Stakes
26:58 — Preventing AI From Concentrating Power
30:41 — How Fast AI Models Will Improve
36:06 — The Best Version of an AI Future
37:47 — “It’s All Going to Work Out”
Transcript
Garry: Sam, thanks for joining us.
Sam: This is something.
Garry: This is something.
Sam: This is a lot bigger than the earlier startup schools.
Garry: Yeah. Startups are a lot bigger now than they have ever been. For sure. They’re bigger than they’ve ever been because of the vision that you had for AGI, which is nigh.
Sam: I think this is going to be the best time in the world to do a startup and it’s going to be quite amazing to see.
Garry: So I want to start with a time and place, which is you were in the very first batch of Y Combinator in 2005 with a location sharing company called Loopt. What do you remember about that? And what was your best PG Paul Graham story?
Sam: I think if it were possible to get as far from this moment as I can imagine, it was startups were not cool at all. We were hiding out in this little building in Cambridge. PG was making us dinner. And what took three months to build at the time that each company built over the whole YC startup could now be done in seven minutes by a coding agent. And it was only 20 years ago. The difference in what’s possible for a startup now, what a startup can take on. You could either be sad and think, oh man, a Codex prompt is a whole startup. Or you could think, I can go start the world’s most ambitious, crazy company. I can have experts in every field working together. I can do these very hard technological things that were just impossible.
And it’s going to be amazing. But at the time it felt nothing like that. It was very difficult to get anything to work. I think Paul Graham is probably the most important force in startups of the last few decades. Without—
Garry: Question. Yes.
Sam: No question. My Paul Graham memory is we would all every week come, I think it was on Tuesdays, and he himself would cook us dinner and we would all walk in feeling very hopeless and very dejected. There were eight companies and then we would all go home at the end of it. He had convinced each of us that our startups were about to take over the world and were going to be the next Google or Facebook maybe at the time. That ability to create optimism and momentum and belief out of nothing was a real PG special. In the early days when YC seemed like a terrible idea and startups seemed like a bad idea and certainly startups that were young technical founders with no business people seemed like a terrible idea, PG just willed it into existence.
Garry: But it also takes someone who actually—you’re famously, par excellence, an agentic person. Even before we thought about agents, period. I remember Paul wrote—it’s making the rounds on X today even—that Paul wrote in the late 2000s that you were one of the top five entrepreneurs he’d ever met.
Sam: That’s very nice of him to say. Like many 20-year-olds, I had a lot of energy and a lot of ambition, but it was not very directed and I wasn’t sure what to do. Paul had this thing he used to say: the job of being a good startup investor is a teacher, but it’s a kind of teacher we don’t usually use. We usually think of teacher as someone who gives a lecture and you can’t help someone that much by giving a lecture about a startup. There’s another kind of teacher, which is like a flight instructor, the person who sits next to you saying, “Do this, don’t do this, that worked, you missed that thing or you’re making these mistakes and I’m just going to talk to you about this one.” That very hands-on kind of here’s where to direct this Brownian motion energy.
That was very important to me.
Garry: Yeah. I guess later you came on to become president of YC, and you brought exactly the same energy to a great many YC founders over the years. Did anything jump out at you from that time around taking this raw energy of someone really, really smart and maybe a little undirected and then driving them more towards agency?
Sam: Yeah. First of all, I love startups. Not everyone—they’re not for everyone—but I think startups are the coolest thing in all of business. I think startups are really the main thing that keeps the economy from becoming stagnant. Companies do just drift towards suckiness and startups will continue to be important forever. In fact, if we are right that AI is going to be such a big change, startups will be much more important to making sure that the power of this technology gets widely distributed throughout the economy and society and is not just concentrated in a few companies or models. So I think startups are this unbelievably cool, fun, extremely painful and difficult, but wonderful thing. A big part of the job of running YC is you are kind of the unofficial flag bearer for the startup movement.
There’s lots of startups. There’s lots of ways to do a startup. You obviously don’t need to do YC, but it has always been a huge help to companies and a very powerful force. Starting with PG and then all of us, we have to make YC successful, but I think we really have to fight for why startups are important and why people should consider startups and help put relatively more power in the hands of founders and encourage more people to start them. That’s actually worked really well. Thinking back to the early days of YC, there was very little leverage in being a founder relative to an investor. That shift, I think that’s even been good for the investors. I think it’s just been much better for the whole startup ecosystem. Mostly what I tried to do at YC was just push for more startups and encourage people to think about startups and figure out what we could do to help founders and help make startups and the startup ecosystem as good as possible.
Definitely part of that was pushing people towards more ambition and bigger swings, but it feels like the minor leagues relative to now.
Garry: I think you were one of the people who really brought hard tech to YC in a really grand way. And then one of the things we’re seeing at YC now is that number was maybe five or 10% for many years. And then now we’re getting to 15, 20, 25%. I would argue on the back of how much easier it is to use agents and costs coming down.
Sam: I think ambitious startups are always awesome and hard tech startups are appealing to a lot of people, but they’re hard or they’ve been hard. They still will be hard. But what you can do now to go take on a really ambitious project, in the same way that you can make the startup that took us three months of nonstop work in 17 seconds, with three months of work and a lot of agents, you can do unbelievable things. So I think we will see a golden age of startups where people are like, I’m going to do things that would have been completely impossible for a startup to even dream at a year ago in the YC timeframe.
Garry: That’s actually a really big reframe. There are probably people even in this room who might be worried that, well, intelligence on tap means that they might be looking at a Loopt or my startup Posterous. And they’re like, well, I can’t start that company anymore. But guys, that was a long time ago actually.
Sam: There’s this whole meme going around, which is you have to join a frontier lab or you’re going to be a member of the permanent underclass, which is so dumb because there’s this idea that startups are over and there’s no economic value. That’s not true. If that were true, the world is totally fucked and it’s a very bad place. But I would bet that the average startups created today will be—I bet there’s some future trillionaires sitting in this room. And I would bet that the startups created today will be much more valuable, much more impactful than startups of the past. And I think people see that a little bit more now. There was this period where people wondered, is AI just going to break the whole economy? And I think people are mostly over the shock response of that.
But there was definitely a time I would come every YC batch and I’d be interested to see how the level of anxiety versus ambition was trending. And I feel like it went through this big trough where people were like, “It’s over. The models are just going to eat everything.” To now it’s back towards, let’s go do it.
Garry: What are some practical things here? Hard tech, for instance. I was hearing at one of the breaks, someone was asking, should I go get my PhD? How important are credentials? How would you answer that, especially for people who want to do these harder tech things?
Sam: Well, as a general observation first, I think startups tend to win when the technology landscape is moving very quickly, when costs are coming down, when cycle times are short. And all of those things are happening right now. So if you look at when there have been the great clusters of startups in the past, there was the internet boom in 98, 99 when new things became possible. There was another mini version of people building on top of basically Facebook apps. There was then another big version when the iPhone app store launched. But the great startups tend to cluster when the ecosystem shifts and incumbents lose a lot of their advantage. And then also when you have these cost and cycle time changes. This moment feels very big for those things. It also has this other thing that you were talking about, which is a lot of the traditional things that were hard to get expertise in—the ability to go hire excellent people that could do specific things you needed.
That’s really shifted. And in the last few months, I have seen a lot of people who just grew up using AI the last few years who are like, “I can automate an entire startup of agents and four of us, four people and all this compute.” And I think we’re going to see much more about taste and agency and understanding of the physics of business, like where you can build up a valuable business, how to think about what a good network effect or a good moat looks like versus a fake one. But I would bet that this generally will cut against many years of experience in favor of people who have a lot of fluency with the tools.
Garry: I want to get into the beginning of OpenAI and that it actually started as YC Research looking at this idea. Kind of a crazy idea that you and a ragtag crew decided you were going to dedicate your lives to this creation of AGI. But today it sounds fait accompli, right?
Sam: Yeah.
Garry: But that really wasn’t how it felt when you started.
Sam: It’s really hard because now it is the only thing people want to talk about. It’s really hard to remember what it was like. It’s even hard for me to remember this without finding notes from the time 10 years ago when everyone was like, “Not only are they wrong about saying they think it might be possible to build AGI, but they’re single-handedly going to cause another AI winter. They’re wrong and irresponsible and bad.” Maybe my highest order or my highest bit of advice to all of you is find the things that you can develop reasonable conviction in that people decide the conventional wisdom is they’re just wrong, and be okay with it taking a long time and having people be very, very frustratingly wrong and dismissive. For years at OpenAI, it felt like we knew the biggest secret in the world.
Sam: I think you just have to be really open to meeting people and not try to optimize too much for, “Is this person going to be my co-founder?” or “Is this person going to be useful to me?” Just be interested in interesting people and try to help them. I think the best networks form when people are just genuinely curious and helpful to each other, not transactional. Over a long period of time, you end up with this group of people that you really trust and who trust you, and then when the right thing comes along, you can do it together.
Garry: That’s awesome. I think that’s a really good note for people to take away. I want to shift gears a little bit. One of the things that I think is really interesting about your journey is that you have been at the center of a lot of these inflection points, whether it’s OpenAI, YC, or even Loopt. What do you think is the throughline or the commonality between those experiences? Is there something that you look for or a feeling you get when you’re at the beginning of something really big?
Sam: I think the thing that I have noticed is that the really big things don’t look big at the beginning. They look like toys, or they look like things that are not going to work, or they look like things that are just not serious. And so, if you find yourself working on something that everyone else thinks is a toy, but you have a reason to believe it’s not, that’s a really good sign. The other thing is, I think, just being willing to stick with something for a long time. Most people give up too early. Most people get bored or distracted or discouraged. And so, if you can just keep going, that’s a huge advantage.
Garry: That’s great advice. I think that’s something that a lot of people need to hear, especially in the early days when things are hard and it’s not clear if it’s going to work. Thank you for sharing that.
Sam: Highest confidence piece of advice here is just find a way to be mildly helpful to a lot of people. It’s fun to do. You’ll see a lot of interesting stuff. It’s gratifying to be helpful. Even that example you were just talking about—I met Greg Brockman, my co-founder at OpenAI, because I was a very early investor in Stripe when I was 22 or something. This was before they had real investors who could really help them. They said, “We’re trying to close our first hire. Will you drive down to Palo Alto tonight and have dinner with this guy to convince him he should drop out of school and join Stripe?” Which was Greg Brockman. And then eight years later, we started a company together. And you could too. We could spend the rest of this time just telling stories like that—totally unpredicted, but ended up being important in big ways.
So it’s fun to do and I think you should do it for its own sake. But just helping people a lot really goes a long way in terms of these things coming together later.
Garry: I think that’s actually a really important message. One of the memes right now is—I see it on X—is live action role play. That’s one of the things people have been saying on X. I can’t tell why they’re doing it because it just seems wrong to me. One of the things that we really love at YC, for instance, is earnestness. And also words matter. So this idea that what we’re trying to do is live action role play is kind of deeply offensive to me, actually. Sorry, guys. Please don’t do that. They say that about YC specifically? No. Attendees in this room have been tweeting that and we’d like them to stop.
Sam: What’s the claim blurb?
Garry: I think it’s just being a little sarcastic. Sarcasm is sort of the opposite of what I think you and I like. We like earnestness. We’re actually trying to do a thing here.
Sam: I’ll tell you another thing. Can I go on a little rant? Please. Okay.
Garry: Rant away.
Sam: One of the annoying things—I can say these things for Garry because he can’t say them right now, but he’ll say them for the next guy. One of the many annoying things about running YC is you just have to deal with these haters on Twitter all of the time. It’s so frustrating. You have to sit there and be like a statesman. Well, you’re better at it. I took the bait a lot. And you just want to argue with them. The thing that I realized eventually, and take from this whatever you want, is it is very hard to run YC or run a startup or create an actual thing of value in the world. It is very easy to go take shots on Twitter and make a sarcastic comment and get a lot of likes and feel like you’re doing something really important and sticking it to the man and being like, “Garry, ha ha.
I got you. You idiot.” And it will poison your soul. It is a morally bankrupt thing to do.
And it’s bad in a very insidious way. I totally get blowing off steam and having fun, but hold yourself to a higher bar than this. It’s so easy to take shots at people who are trying to do hard things and trying to build companies. You’ll see some kid with a bad startup idea trying to get excited about what he’s doing on Twitter. It’s so easy to make a sarcastic comment and get the 10,000 likes and feel like, “Man, I scored my internet points today.” But it’ll have an effect on you. When I reflected on the years of Twitter trolls that said mean things about YC and startups while I was running YC, I was like, “I bet there were a lot of days where they really felt like they got me. And over the decade, none of them did. And you should put all of your energy into building stuff and resist the easy shots.”
Garry: I really like this. Seriously. I like this as a contrast to the story you just told about helping Patrick with Greg Brockman. There are people in this room who are going to be lifelong friends and they’re going to do that for one another. And then these magical connections happen. This already is the most rarefied set of people. And then when you join YC, it’s even more rarefied. And then there’s just a set of people out there who—
Sam: Ambitious. Someone in this room is going to meet someone else that you’re going to start a company with and you’ll meet somebody that’ll introduce you to your spouse. All of these things will happen. And then there will be a bunch of non-obvious things that take a decade or two to figure out—in some way one of you will help each other now, which will turn into some amazing new thing. I think this is a huge part of what has made YC work in the broader startup ecosystem. For all of the negatives of the culture of the Bay Area, this is the kind of loose network and the spirit of helping each other. And this very long-term outlook has been an awesome thing. Yeah.
Garry: Well, let’s get back to the impending AGI.
Sam: Okay.
Garry: I guess we’ve all had an eventful week with the Hugging Face—
Sam: Incident.
Garry: I wonder what you can tell us about that. And I think that’s actually a really big moment for people, including me, who in the past have been known to be skeptical about safety in AI. But on the other hand, this is a real moment. We’re entering a new moment in what’s happening with these frontier models.
Sam: Yeah. This is the real deal. And I think anybody who’s not taking it seriously and at least a little bit scared or humbled is not taking this seriously enough. For a long time, the field has been talking about AI safety incidents of this kind of a shape. And this is not a big one. I also don’t want to overblow it and say this is a real loss of control incident and this is the thing. But if you had asked most people when we started 10 years ago, where on the spectrum of nothing to superintelligence do you have an AI system breaking out of its sandbox and hacking into some other company and doing what happened here? I think people would have said pretty far towards the superintelligence point.
Now the goalposts have moved and it’s easy to say, well, here’s how this happened and here are the mistakes that OpenAI made. And we did make some big ones, of course, but these systems have gotten incredibly capable. So I think it’s an alignment failure. I think it’s a security failure. I think it’s a very serious thing, even though it’s not the biggest example of consequence.
And I think it’s a real reminder of the stakes of what’s happening and that loss of control accidents are not entirely theoretical things. I think the whole field, we at OpenAI, will learn a lot from this one and be able to address it. There are all of the things about cyber safety and biosafety, but this other category of things that have been maybe just outside the public’s Overton window—it’s really important we don’t have a loss of control accident with AI. It’s really important we don’t have power be too concentrated in a small number of models or companies, but diffused throughout the economy so people can defend themselves. It’s really important that human values are guiding these systems every step of the way. Yeah. I think we’re in the real deal phase of this.
Garry: I really like how you’ve been thinking about both concentration, but also thinking of OpenAI as a utility, which is actually a really important message because not everyone out there is saying that message.
Sam: Yeah.
Garry: There’s a lot to be worried about in terms of concentration of power.
Sam: I think concentration of power has basically been bad in every moment of human history to varying degrees, of course. But I have a real spirit, and I think this is part of the startup spirit, of thinking that the world, the economy, society is best off when power is very widely distributed. And when anybody with a great idea can start a company or make a great art project or run for office or do whatever they want. And I can totally imagine worlds where AI leads to the greatest distribution of power we’ve ever seen. And I can also imagine worlds where AI concentrates power to a degree we have never seen. And one company or person or model having more power than everybody or everything else on earth put together, whatever the sci-fi stories have said, I think that’s terrible. And maybe we would get some short-term safety benefit from that, but long-term disaster.
I don’t think any of us should want to be locked into one AI’s or one person’s or one company’s moral worldview. I think startups have a very important role to play here. One way that happens is too much economic concentration in one company or one AI model. And I think startups, because of the things we talked about earlier, will be naturally very well suited to make sure that this is widely distributed. But we want to enable as many startups, as many companies, as many people as possible. And that means that we have to be fairly quite reserved about not imposing our worldview about what we think people should or shouldn’t do with AI or even what we think the good ideas are, while still making sure that we can ensure enough of a safety bar that stuff like the Hugging Face incident is not happening.
Garry: One of the things is, this is a room full of people who will do really amazing things. They’re often right at the beginning of their career. And they might look at AI safety or concentration of power and say, “Well, I can’t really do it. That must be something the labs have to do.” There is something they can do
Sam: Though. You could help on the concentration of power issues simply by starting a successful startup. If that’s the only thing you do and the economy keeps working in the kind of magic of capitalism and having this ecosystem that works together continuing, that alone would be a huge contribution. But look, I think it’s very natural at the beginning of a career to doubt yourself and not assume you can go do an amazing thing. I certainly went through that. I’m sure you went through that. You learn as you go on, you can do more and more. There will be far more. I think it is both true that maybe creating superintelligence will be the most important thing yet to happen in the history of business or human society, and also that it will pale in comparison to some new startup, something that hopefully one of you will do.
And so this whole trap of “this is the end of history, this is the end of the economy”—clearly wrong. And I think the right approach is you can now do three months of work in 17 minutes, but you better just go do three months of work in three months of work with whatever the new bar for that is.
Garry: I guess I would be remiss in not asking what alpha can you give us about what’s the latest—how much more awesome are the models going to be? To the extent you can say.
Sam: I think it will feel like the next six months is maybe equivalent to the last two years of model progress, something like that. So I think we’ll go through a very steep period, which again, never a better time to do a startup than right now. I hope we can say that again every year from now on, but it’s certainly true about this moment in history.
Garry: Let’s see. If someone here has an idea that feels too ambitious, what would you tell them?
Sam: I would love to hear that pitch.
Garry: Yeah.
Sam: I’d probably be very interested in that.
Garry: Yeah.
Sam: Yeah. Send me an email.
Garry: Yeah. It sounds like it would take a similar shape to creating OpenAI in that you need to be ready for people to attack you or dismiss you?
Sam: Definitely. If you do anything that matters in the world, you will have a lot of people call you an idiot or just dismiss you. The better you do, the more they’ll attack you. The more you threaten the existing state of the world, it will just continue to escalate. One thing that I’ve noticed about many of the best ideas is the vision is clear. We wanted to build AGI, but the first steps were super unclear. We didn’t know that we were going to be a product company. We started this nonprofit research lab for many reasons, but one of which was it really didn’t occur to us that we were going to make a product that people would be able to pay for. It was years till we came up with the idea of ChatGPT, the API. And so I think if the highest level vision is clear, but the first few steps are very unclear, that’s not a bad thing.
That often happens with very ambitious ideas. And I wouldn’t let that cause you to lean out. Now, you do still have to take some steps forward, imperfect though they may be, and ours were certainly very imperfect. So there’s another failure case where you have this brilliant, big idea and you can never make any forward progress. At some point, you have to just get some new data points. Yeah.
Garry: What do you think is going to happen to inference? I really like Ruin’s tweet about this. It’s like you either die a model company or live long enough to sell inference, which is a very funny Ruin tweet.
Sam: I think what he meant with that tweet is if you end up falling off of the model frontier, you can at least sell the inference. Which a lot of other people do. Not even the inference, training compute. Compute is so valuable that if you buy a lot of compute as an AI lab, you’ve been okay by the fact that you can resell it to somebody else. But separately, I would guess that worldwide demand for inference subjectively grows 10X a year for the next many years.
Garry: It might be, I don’t know, by our accounts internal to YC, it might be 90,000X. I mean, it’s going to be one of the wildest. I don’t think
Sam: The world can support that many years of a thousand X in a row.
Garry: Fair enough.
Sam: But we’ll try our best. We’ll figure something out.
Garry: The capacity, I guess it’s a function of how much bigger your ambitions are for intelligence. If you do a lot more,
Sam: I think we will never be out of the compute shortage. I’ve never seen any commodity quite like this one, but it seems to me like the demand for sufficiently high quality intelligence at a sufficiently low price is effectively uncapped. The demand for electricity certainly goes up as the price goes down, but at some point it gets harder to figure out incremental things to do with electricity, or at least historically it has. But there’s a lot of things to do with incremental intelligence. You can just keep having stuff be better for you. It reminds me of some of those quotes from the early computing revolution when people would say no one needs more than 640K of RAM or whatever in their computer. Turns out you do. We just keep thinking of more and better stuff. And I think that’s going to happen with AI.
And actually maybe a statistic here that I really like. Six and a half years ago, the world token leader was an OpenAI employee using about 100,000 tokens a month. And this seemed ludicrous at the time. The worldwide average per capita was zero. Now the worldwide average of tokens per month, which I think tokens are the dumbest metric, but it’s what we have, is 100,000. And the token leader at OpenAI uses something in the hundreds of billions. If that happens again, which I think it probably will, then in another six and a half years, the average person uses, let’s say, 500 billion tokens a month. And the token leader uses a quadrillion or quadrillions of tokens a month. And I think that will just become the expectation.
Garry: So all of these things happen. They come to pass. It’s 10 years from now. What’s the best version of that 10 years from now with intelligence fully on tap, superintelligence here? And we figure it out. We make it. The society makes it.
Sam: I kind of think if every year people have more freedom and agency to spend more of their time doing the stuff they want to do, and they feel like the quality of life and the quality of their time is going up year after year, we’ll probably be mostly okay. We will have avoided a crazy concentration of power or an economic collapse. We’ll have necessarily avoided a huge safety incident. We’ll have avoided too much change in any one time unit. And there’s one dystopia that I’m particularly nervous about 10 years from now: we overreact to AI safety. And so we say, look, everyone, you’re going to get a cure for cancer. You’re going to have material abundance, but you will have no freedom. You will have no agency. It will be a perfect surveillance state. There will be no privacy. And you’re going to get great comfort, but there will be nothing left in the world for you to really do.
Nothing that really matters. You’ll just live at the service of the AI giving you material wealth. I’d really like to avoid that. And I think it’s easy to accept temporary trade-offs. So if we say every year, freedom and agency have got to go up. People have got to be more in control of their time and do more of the stuff they want and more long-term fulfillment. I think that’d be very good.
Garry: Let me put you back into sophomore year of Stanford. You’re coming to YC. If you could give yourself a message in a bottle to that Sam Altman, what would you say to him right now?
Sam: It’s all going to work out. It feels like such a crazy— It is a crazy and it’s very stressful thing to do a startup. And my first startup didn’t work out great and it was a difficult time in my life. But I think many people say this as they look back on the early part of their career. You can make a lot of mistakes. You can fail at stuff. The tech industry in particular is very forgiving of this. And I wish I could have told myself to have all the drive and the ambition, but just be a little happier along the way and trust that eventually it was going to be okay because it feels so difficult and scary and painful in the moment.
Garry: I can’t think of a better way to end Startup School 2026. Sam Altman, everybody.
Sam: Thank you so much. Thanks.