Why I Backed OpenAI, AGI, & the Zero to One Team | Vinod Khosla

The best founders and CEOs I know always ask me, what can I do better? Not what am I doing great? Let me tell you about that.

What does a great investor look like from a founder's perspective?

At different stages, you need different kinds of investors. The first thing I'd say is in the 40 years I've been doing this, not once have I called myself a venture capitalist. I always say I'm a venture assistant. The idea is if you've done a couple of startups, you can mentor startups.

You had a great hit with Juniper, and obviously KV has been a top-tier firm for the 20 years. Do you care if you're wrong or if you're kind of like not yet right?

There's actually a Harvard Business School case. It starts with a quote from me, "My willingness to fail is what gives me the ability to succeed." I would posit that if you're not going to take risks, you're going to do things that others are doing. Founders who think from first principles almost always do the best if they're trying to dramatically change the rules of the game.

Hi, everyone. Welcome to SPC. Very excited to welcome Vinod to South Park Commons. And I don't think Vinod needs that much of an introduction. I think he's kind of been a main character across all of technology and all of Silicon Valley for the last 40 years. So let's give him a round of applause. Let's get this started.

Thank you. You know, it feels like I'm just starting.

There we go. There we go. Yeah, I think I was going to say, I don't think Vinod needs any help making it spicy, but let's try to have some fun here today, okay? Vinod, I'm going to perhaps start off by asking you to rewind back to not Sun Microsystems, but you had two companies before that. Is that correct?

I had a— yes, two companies before Sun.

Okay, so maybe talk us through, I mean, at SPC, We're all essentially in varying degrees of the minus one kind of like I would say journey, trying to figure out which ideas are worth tackling, trying to gain conviction in an idea. So I'm curious, like back when you were starting off and you were kind of like trying to figure out what to dedicate your time to, how did you go through that process?

My first startup was called Daisy Systems. It was a CAD company, actually started the whole field of CAD practically. Before that, there was graphics that was meant to be CAD, but if two lines crossed, it didn't mean there was a circuit connection. There was no database under the graphics. So it was a direct paper translation. That was quite successful, went public. I did that right out of business school and that worked pretty well. But one thing I realized is you have to have the right teams with more importantly, the right values. And I didn't share values with either the team, especially the investors there. Fred Adler was our investor. I don't mind naming names, uh, but being values aligned is really, really important. So I left and then started looking at two different ideas, both with Scott McNeely. One was a temporary sort of, it'd be fun to do this kind of idea. And we sort of did that in a very bad way. That failed. But we really also focused on what is the core technology idea. And I found Andy Bechtolsheim and I decided I wanted to work with Andy. By the way, one of the really important lessons Andy was developing the Sun workstation at Stanford. Sun used to stand for Stanford University Network. So, um, that's where the name came from. Uh, and he was licensing this, his design, because he was a graduate student, for $10,000 a pop. I said, and this is a really important lesson, he had licensed it to 6 other people, which in those days $60,000 for a grad student was a shitload of money.

I think it still is a shitload of money even with inflation.

Yeah, but think of inflation adjusted from 1980 dollars. But I said, "No, Andy, I want the goose that laid the golden egg. I don't want the golden egg." And in the end, he became a co-founder, which took me a lot of time because I had to convince him to drop his PhD. Then we went over to Berkeley and convinced Bill Joy to drop his PhD. I was good at selling, and you do lots of selling, but the team you build ends up being the company you build. When most people said, "Well, you can license the technology and you'll be fine." In fact, none of the other players, and there were 6 others, including Jim Clark at Netscape, who had done Silicon Graphics, licensed the same technology but never came to much. Because one principal lesson I've learned across the years, the team you build initially is the company you build. And SPC does a great job. In fact, this is the one piece of advice I keep giving YC. They don't do a good job of helping build teams. And you're doing essentially that.

I have no comment on YC. Um, but you know, I'd say that, but maybe I think, how many of you ever have used a Sun terminal? It's kind of amazing. So I went to Carnegie Mellon. So back in, I got there in 2000, I was there for 4 years. And all our terminals in the computer labs were basically, well, the good ones were Sun. You could also use the Linux machines, but at that point they weren't that great. And I think it's kind of hard to understand today just how good those machines were. And then obviously if you've And again, the kids today don't use Java, but Java was kind of like the programming language that all of us were taught, like, as being essentially the gold standard back in 2000.

And by the way, there's an important lesson there. We had decided— we are much smaller than almost anybody else in the computing business. So the most common piece of advice I got was, why don't you build terminals for DEC VAXes or something like that? And I said, I want the whole enchilada. I don't want just the trimmings. Uh, so we went for the more ambitious goal because we had the right team. But one of the things we did is to gain mindshare because you're a small, tiny startup. And in fact, Apollo was a workstation startup much larger than us and with fancier management, and they had martinis at lunch. Uh, uh, we had McDonald's. We decided to look for unusual leverage and wanted to penetrate every computer science department in the country in those days. And the reason I tell this story, because many of you are at this stage, we could talk forever, um, finding unusual leverage is a key to a startup. And we said, we want all the universities to be using Suns and identified the 50 computer science departments we'd be in, made a rule, you, nobody could travel without visiting a university university that was closest to where they were going. And in the first year alone, we got 42 of the 50 computer science departments we had targeted. And it became an unusual source of credibility because wherever corporate people went or whoever they hired, they had heard about Sun.

So, I mean, how did you figure that out? Was that simply like, where do you get this moment of insight?

Well, our competitors had a lot more money, a lot more advertising. We didn't have money for any of that. So we had to get creative.

You know, Vinod, it strikes me, and this is a theme that I think we will keep on coming back to, but you are in some ways consistently 10, 15 years ahead, not just in terms of the actual technology, but also how that technology in some ways is brought to the market. And what I think about is that it's cool to start a company today. I hope a lot of you start companies, but in 1982, that's still like, I don't, I can, I can't imagine that many, it was that popular a path. Right. And then I was going to get to next, like in 1986, when you became a VC, not everybody wanted to become a VC either. So I guess the question here is, um, All of this stuff sounds very cool in retrospect, but at the moment when you're kind of like alone on the island doing this by yourself, what gives you in some ways that early conviction? Because I think that is a question that a lot of people at SPC are asking themselves, which is how do I have conviction that this shit is right?

There's two ways to think about what to do. One is conventional wisdom. And if you're extrapolating the past to predict the future, The big companies do that very well. Intel knows how to go from a 15-nanometer process to 10 to 7 to 5. Um, and, and big companies do well in extrapolating the past to predict the future. Entrepreneurs fundamentally have to invent the future they want in a very different way. And you have to have conviction about that. Now we tend not to do a lot of, we are slightly better, slightly more efficient, slightly lower cost. Uh, the risk of that kind of startup is very high. Uh, if you're fundamentally reinventing something and changing the assumptions, then you have a lot more time to get it right and a lot more iterations you can do. So, uh, in— I'll tie this back to if you get the right team, it'll iterate enough. It doesn't even matter if you got the spec right. Iterate to the right answer. And almost all good startups do that. Uh, and, and this is why the team's more important than the market you pick. If you picked it wrong, the right team will iterate to the right answers, especially if there's a culture of, uh, debate and disagreeing. If there's a single founder and you're dictating terms, then you're not going to evolve. Great talk, TED Talk called Dare to Disagree. I'd recommend reading it or watching it. And that woman, Margaret Heffernan, has written a couple of books. So how you set up a culture where no matter whether you're senior or junior, you speak up if you have a concern or an idea and you detect problems much early. And then as a group, you can evolve your direction. Pretty important.

Yeah.

By the way, one of the reasons Andy Grove at Intel essentially started dress down, no given parking, the original culture of the Valley in the '70s was he wanted an engineer, whether they were junior or senior or a VP, to speak up and not be intimidated that the executives were all wearing ties and coats and Then they didn't have ties and coats, and so they couldn't speak up.

Now let's make it a little bit spicy. So do you think— I've actually never worked at Intel, but I've heard legendary stories. Were they— you know, I think the Valley has lost a little bit of just like that sharp edge, that sharp edge that actually makes companies better. Like, I think that— do you think that we have lost some of that edge that Intel had, whether it be through We have to potentially let go of the bottom 10% of the company to essentially aim higher. And actually, this shit is competitive. We are in the most competitive industry that has ever existed.

So most innovation happens when a founder is running a company. And there's a few sort of exceptions, but not really. Founders who think from first principles, as opposed to use historical things or what works or does almost always do the best if they're trying to dramatically change the rules of the game, which is when you build big companies. And this is why incremental stuff is hard, is very hard. I think that's still a characteristic of the founder. Now, take a couple. I'll give you my two favorite examples. Examples. Um, so we invested in Larry and Sergey, what year was it, '97 or '98?

Yeah.

Um, they didn't care what anybody else was thinking. If they wanted to start a self-driving company, they started one. It made no business— no business school professor would say start a Waymo. Uh, my favorite example is Jeff Bezos as a retailer, and we invested in him in '96, uh, when I was at Kleiner. Um, he didn't care that he wasn't a computer company. He started AWS, but he was thinking from first principles. Uh, you'd think IBM would do that or somebody like that, but they didn't. And so The having the moral authority of a founder and always having this questioning attitude and asking a lot of questions is in fact the key to evolving a strategy of a company. Almost certainly for those of you who sit here in 2030 and are successful, your plan will look very different from where you start today.. But if you have the right team, you'll hopefully have the same team.

Yeah, I mean, I think that maybe to talk a little bit about the SPC investing philosophy, right? Like given that we are not even necessarily at the pre-seed stage, but we are really investing our time in founders in the minus one stage. The, the only thing that matters for us is the quality of the founder and the team. Right. And I think we obviously need to be passionate about the market and the product that they're building, but ultimately that stuff changes. But we need to have a ton of conviction in the founder's resilience, brilliance, grit, and just ability to have the passion to build something in the asymptote. But let me ask a question here. So one of the things I tell founders is that if you are essentially building something that is 10x better on some dimension, and in some ways you're embodying that kind of like 10x betterness, then there's likely something else that you're not good at, right? Because I haven't met anybody who is kind of 10x good at like everything, right? Would you agree with that?

Oh, absolutely. You know, and one of the things I advise founders, and you'll go out at some point, look for a full series venture funding. It is better not to have a complete team than have the wrong people in spots, you know, so common. If you want a business person, go hire an MBA. Really bad idea, right? Uh, you'd rather get the help of somebody who can judge the functions you can't judge to help decide who to hire. So I'm completely fine with incomplete teams. I much prefer them to mediocre complete teams. And usually a founder who's excellent at one thing, let's say AI algorithms, they're not going to be the best at saying who's a good marketing person or a product person or a CFO or a salesperson. And they almost always do the wrong hires. And so be really, really careful about that. We keep coming back to the team you compose. And that's what SPC has done well.

Thank you. And you have advised like, you know, hundreds of founders over the years. When somebody is kind of in some ways starting off on their ideation journey, are you a fan more of top-down? So like, hey, go identify markets you like and then kind of systematically tackle them. Or are you much more bottoms-up, like go start building, go start tinkering and see what kind of arises from an emergent property?

You know, there's a time and place for both.

Okay.

I like to give the following analogy. If you go to a British roundabout, I don't mind going in circles, peeking down roads to see which road to pick. Once you pick a road, you're going to end up where the road leads. So better pick which turn after circle, roundabout you pick. Boards, by the way, generally hate this. Yeah. So I'm always arguing with boards about it's too early to pick a direction, like let them explore. So there's a clear exploration phase. The key to that is really low burn rate till you decide, and then you can start executing down the path. But if you're going down the first path you think of, you're not necessarily optimizing for that or for the team you have.

I mean, we talk about this at SVC often. I think Jonathan often says this, which is it is unlikely that the first idea that you have is the best idea you will ever have, just by math. Okay. So even if you give it a little bit of time, like, you know, give yourself like a month, 2 months, and then even if you come back to that first idea, at least you know that you're kind of comparing it with a few other turns in the roundabout, right?

Yeah, absolutely. And so I do think there's a really important exploration phase for a company that generally is ignored and mostly not done because you get— I hate 5-person companies focusing on ARR. Like, I'm doing 100 ARR. Well, it's $10K. You could get your father to buy $10K worth of stuff. Focus on where you want to go in 5 years and then work backwards to where you want to start, because you have to do both. And be willing to change your mind and iterate. One of my favorites I advise people is don't make a plan, but plan to plan. What does that mean? Every 3 months you can change your plan or tack right or tack left like in sailing. What's important about making a plan is identifying the variables that you think are currently important. And as you learn more, new things get added to variables that are important in your head and some get dropped and you iterate that way. And this is where having the right team, having the right culture of debate is so important. If people in your team think you know it all and are afraid to disagree, you're not going to get that input, especially early on. Okay.

Vinod, how do you think, before we kind of move on to AI and AGI, maybe one question around how, what should founders, what does a great investor look like from kind of in terms from a founder's perspective? What should all of the founders in this room expect from a world-class investor? Somebody that— what should their— what bar should they hold their investors up to?

Yeah. At different stages, you need different kinds of investors. So to be clear, the first thing I'd say is in the 40 years I've been doing this, not once have I called myself a venture capitalist. Not once. I always say I'm a venture assistant and 25 years ago when I was a client at John Doerr would always get mad at me for saying I'm a venture assistant. But the idea is if you've done a couple of startups, you can mentor startups and assist them in the things they don't know. And almost certainly great founders, uh, the majority of the time are incomplete in some dimension. And so your job is to help them with that and challenge them with their strategy. I hate the culture. There's a number of firms in the Valley who push the idea of we are founder-friendly, which, which really means they're never going to challenge you. They want to be popular with you as opposed to help you. And I always draw the analogy. It's like my kids. If I always say yes, they'll end up really spoiled.. And you mostly— how many people have kids? A few. You mostly, you want their welfare at heart, but you want to teach them what's right and wrong or debate them. And over time, mostly pass the decisions to them. I have a rule with my kids, they're all grown up. At age 12, I made 90% of the important decisions and they made 10%. And by the time they were 18, they made 90% and I made 10%. It's a pretty good rule for kids, but startups are very much like kids. You don't always want to say yes to everything and too many VCs are in the popularity game. And so they're never honest with you. Um, so. There's a saying on my website that we put there in 2004 when we started the firm, that we prefer brutally— brutal honesty to hypocritical politeness. 95% of what feedback you'll get will be hypocritical politeness, and it doesn't help you. It makes you feel better, but it doesn't really help you. Real well-intended critique of your strategy is what you want. And people who challenge you think about others and can then otherwise assist you in other ways.

Yeah. I mean, I think that's really well put, right? And I think even as founders, one of the pieces of advice I'd give is almost just set that tone with your VCs or your venture assistants, which is that it's not that helpful to send an investor update and to get like two words back, great job crushing it on ARR this week, you know? And to actually kind of be like, what are the 3 things that you think we are, we could be, you know, like we're not attacking or like what are my opportunities to 10x from here? Because in the startup game, if you're not 10xing like on a regular basis, this is not like, you know, most startups are not about 20-30% gain, you know, quarter over quarter or like year over year.

Yeah. So one other thing that I think ties to this, I don't go to board meetings anymore. 'Cause frankly, at the board, everybody, every board member is trying to look smart as opposed to say what's really helpful to the founders. Boards have really bad dynamic of, and then if one person says something, others want to disagree with it and you end up with groupthink. The best founders and CEOs I know always ask me, what can I do better? As the most important question they ask, not what am I doing great? Let me tell you about that. All right. So what's not going well? I was just talking to Amjad from Raplet, one of our companies this morning.

Also an SPC company. Huh? Also an SPC.

Yeah. My first comment to him, tell me what's not going well. Right. Then we can discuss what's not going well and solutions to that. As opposed to, he could talk about all the things, lots of things going well, and we could have filled up the hour. I'd feel good, he'd feel good, and we'd go away, but I'd have added zero value to his thinking. So really important way to the culture and strategy. To your question of VCs, most VCs can't do that. And what— and I tell this to younger people in our firm, you haven't earned the right to advise an entrepreneur if you haven't started your own company. You don't know how hard it is. You don't know all the emotions. I have an '86 presentation on our website that I first did in 1986. I called it the entrepreneurial roller coaster, where the highs are high and the lows are really low. Like, that's the tagline of the presentation. Until you appreciate the dips, you can't help an entrepreneur.

Yeah, I mean, I was just reflecting that, you know, I think of what we're trying to build at SPC also as a startup, and it's still insane to me. Like, you will wake up in the morning and everything's going pretty well. You get a few emails that make you happy. By midday, you're just like, this is shit. I'm not building anything worth building. And then by the end of the day, you're still feeling pretty good because you're still alive, you see another day, right? And that roller coaster, it's amazing. But Vinod, let's move on to maybe AI. What did you— I mean, how did you invest in OpenAI back in 2017, '18?

Yeah, 2018 is when we made the decision to invest.

And like, I mean, I'm Makes no sense. Makes no— What gave you, in some ways, you can call it foresight, courage, delusion to make that investment at that point?

So I think it's very important to have a belief system. What people do is believe what others are believing as opposed to from first principles, think through. So I'll give you two examples. Let me start somewhere else. 1996, you know, Sun adapted TCP/IP in 1982, only once. Nobody else adapted it. Fine. By 1996, 1995, it was clear to me TCP/IP was scaling. I talked to Cisco and their CTO told me they'd never do a TCP/IP router above OC-12. Talked to Lucent, Alcatel, nobody was doing it. I talked to AT&T and all the providers and they were not ever planning on using TCP/IP. I called Pradeep and I said, let's— we worked together for 6 months building a plan to say we will build an industrial-grade router for the public network. If you look at any literature from 1996, ATM was going to be the asynchronous transfer mode, to those who don't keep up with telecom, was going to be the protocol of the internet. In fact, I consider this one of my biggest contributions. We said, "Well, everybody's going on ATM. We will only do TCP/IP." And Cisco was doing multi-protocol routers. So they were still routing AppleTalk in 1996 and every conceivable protocol. We said, "We'll build a TCP/IP only router." for the public network with a set of characteristics like atomic updates, like rollbacks, like almost like a database kind of functionality, but in network routing. And we clearly focused on this. Nobody else was doing it. And guess we won. And I don't think TCP/IP would have, would have become the protocol for the internet if we hadn't started Juniper to do this. It's massive. But the result of that was we have, on my first $3 million investment, we made 2,500x. You can compute the numbers.

That's what I'm doing in my head, but yeah, $7.5 billion. Okay.

All right.

Yeah.

Right. Uh, in, in fact, it was either Pinterest or Snap going public, the Wall Street Journal did an article on the highest returns ever, and Juniper was number one in terms of multiple of money invested. In those days, $7.5 billion was a lot of money.

I think it still is.

It still is. It just, there was no outcomes where a venture firm made a billion dollars. Now it's more common.

More common.

But it's having a belief system and building it. But what, uh, to your question on AI, which I want to get to, I saw TCP/IP growing exponentially and near the knee of the curve. And I said, I don't care what anybody else is saying. The, the curve of TCP/IP traffic is increasing at an exponential rate and exponentials look pretty flat in the beginning and then a big curve. Same thing with AI. I looked at, in 2000 is first when I mentioned AI in an article with the New York Times. I literally said in that article, I didn't want to, I wanted to be vague. AI will have us redefine what it means to be human. You can look it up, my interview in the New York Times. And I think we are very close to that now. But 2012, I wrote two blogs called Do we need doctors? Thesis being AI will be the doctor. And then do we need teachers? AI will be the tutor. Both problems we are working on. But I saw the influx of talent into AI. I saw the rates, the beginnings of the exponential in terms of capability. There were all these curves about human performance is up here and AI is down here, but it's increasing. And it was clear to me, if you followed the data, the curves would intersect. If you followed the press, you'd believe nonsense about this won't happen. So it's fundamentally thinking from first principles, follow the data, not what everybody else is saying. If you follow what everybody else is saying, you're going to do the same thing. It's not going to be that distinguished. But this idea of Having a belief system and trying to follow data.

So in that case, I mean, once you frame it, well, once you put it into that framework, I'm assuming that investing in OpenAI in 2017 or '18 must have in some ways actually felt simple.

It was simple because you had been working on it for a while. But sounded incredibly ridiculous. So in the 20 years that Vinod Khosla Ventures has been around, it's the only investment in which I've ever sent an apology letter to our LPs when we made the investment. I said, sounds weird, it's a nonprofit, there's no business plan, there's no product plan, there's no revenue plan, but we are making this investment because we believe in AI. And it was twice the largest investment I've ever made as an initial investment in any company in 40 years. But It's about belief systems.

It's pretty cool.

Are you—

I mean, when you're that early, you're bound to be wrong. You're going to be wrong about a bunch of stuff. Does that bother you?

No.

Like, do you care that, you know, you might— I mean, you had a great hit with Juniper, and obviously KV has been a top-tier firm for the 20 years, but it's also probably gone through its own roller coaster at some times, right? Like, do you care if you're wrong, or if you're kind of like Not yet, right? Maybe—

So there's actually a Harvard Business School case where it starts with a quote from me, why willingness to fail allows me to succeed. I would posit that if you're not going to take risks, you're going to do things that others are doing because that's the low-risk path. And then you're going to do something undifferentiated and marginal. If you're trying to truly innovate and run away, You have to start with a set of beliefs that are different than what others are doing and then find clever ways to hack your way there.

I mean, honestly, like, I cannot repeat this strongly enough, like, as all of you are trying to go raise capital, like, this is actually a question you can, like, you go through your thing, which are what are 3 things in my tech that maybe I am one of the few people in the world that believe to the extent, you know, that like is required. Hopefully all of my friends disagree with this because it sounds kind of wacky and loony. But if you're asking all of your friends and they're like, oh yeah, this is a great idea. I think you have a problem, like especially at the super early stage.

If everybody agrees, then you're seeing consensus. And consensus is a bad idea. So we in our firm don't require consensus or even two people to agree on an investment. If one person believes it, that they should make the investment, they'll make the investment even if everybody else disagrees. I hate an investment, Shwen and our firm can make the investment because we are optimizing for people who are thinking out the box and I may not believe that or— Yeah.

I mean, this is basically what we do also. We only require one person to have strong belief. To kind of make an investment for the exact same reason.

Most of the venture firms operate differently.

Yeah. They have these crazy kind of like scoring rules, like out of like 9.

Yeah.

It's, it's crazy.

Yeah. And I would give you one piece of advice. Most VCs will steer you along conventional lines because the formula works like where it has worked. And that may be actually bad for you. So be careful about listening too much to VC advice. We invested in two things, the three things in 2018. One was OpenAI. The second was fusion, Commonwealth Fusion. Everybody told me it's more than 50 years away and there's no way you can fund this and it's only countries that can do this. Not right. The third thing which is not yet popular, we invest in a public transit startup. So all I'm saying is thinking outside the box is really important. By the way, nobody still believes we can do a public transit startup, despite the fact that we bid on 4 projects where we were not invited to bid, 'cause they invite Siemens to bid on public transit, not a startup. We won all 4 bids. We've not lost a bit yet. It's so compelling, but people still, even other investors don't believe. So thinking out of the box is really important is what I'm saying, and have your own belief systems.

So what are you seeing at the forefront of AI progress right now that you think people in this room probably are also still discounting?

I think we are effectively there already. It's just this future isn't evenly distributed, as they say. For people who are skilled in the art, O3 is probably better than most humans at most things already. But more importantly, O3 hasn't started designing O4, and that will accelerate the rate of development. So how many people have read a paper called Orders of Magnitude or Situational Awareness by Leopold. Yeah, yeah. Right. I think he's right. And most people just can't extrapolate exponentially and have a real problem. But I think we are there. But are there new things that are not proven? I'll give you a trivial example, which was DeepSeq. Everybody just assumed only the big companies can do LLMs, and they proved something different. The, you know, and it's controversial what they really did, whether they stole. It just came out recently that when they say 6 million training run, that was only for the last run. I think they stole a third of the IP. There was a third of clever hacks and a third of real innovation. So give them credit. So I think the most important thing that did is opened up the idea that it's not only the big players with a billion dollars to spend on a training run who can innovate. My favorite is a company we have called Symbolica that uses— any mathematicians here? Probably a bunch. Uses category theory to approach AI, which would be orders of magnitude more efficient than LLMs and transformer models. These things are worth trying because the consequences of success are so large, even if the probability of success is lower. And so it's my current favorite. On how you make LLMs much more, much better, how you embed safety into it. Because explainability, if when you use math, explainability gets built into the system and then safety becomes an easier problem. So I was just talking to at NeurIPS, Yoshua Bengio, and I said, my view is safe math is the approach to add safety to LLMs. And he completely agreed, most likely bad. So lots of things to try still.

So like, how would you prepare your kids for AGI in a world where that actually exists, where there are these machines that are probably better than humans can be?

My favorite question. You know, today you go to any school and you, a kid asks, what should I do? And they say, follow your passion. And I always say that's nonsense advice because do it, follow your passion. If you don't care about supporting your family, paying for your kids' college, maybe paying for a mortgage, then you can follow your passion, be a surfer. And if you're happy with a tent on the beach as your principal living space, great. I think that will change. Kids will be able to follow their passion 20 years from now. They won't need to make a living because there'll be enough abundance. First, AI will do so much. There'll be no difference in the value a farm worker adds or an oncologist adds because they'll both uplevel to the level of the AGI. So how you pay them? Hard question because they have the same value add.

Same value.

It's funny, just this weekend we were looking at the NESTERIC Foundation model in developmental biology, one of my favorite areas to look at. My chief of staff, who's a cancer immunologist PhD, prepared an 18-page memo, which I dutifully read. And then I said to her, "Do me a favor and see if you can use O3 to do a research report on which indications they should go." after with this model. And it produced almost the same document.

That's insane.

And she said, I wasted all this time.

Wow.

So this is becoming very, very real. It just— most people don't realize what's possible.

I mean, I asked—

I'm an optimist. I do think kids— AI will free humans to really pursue what they want to pursue. And you won't have to worry about making a living first and then pursue your passion.

Listen, I'm feeling pretty happy.

This is a very optimistic picture.

I'm feeling pretty happy that my 8-year-old knows how to surf and ski. Okay, so he can be a surf bum or a ski bum. It's up to him. But I'm actually genuinely curious about this, Vinod. How should entrepreneurs in this audience think about opportunities? Because I think a lot of the stuff that people are thinking about today will also be pretty automated by these reasoning systems in a 24-month horizon. I think the world looks very different 2 or 3 years from now.

So I'll tell you about my favorite YC joke, and I'm serious about this because we talk about it. Between starting a batch and 70% of it is AI and ending a batch, feature releases on major models have obsoleted half the business plans by the time the batch finishes. So thinking through where you can put proprietary walls, we spend an incredible amount of time doing that and then guiding people in the right place. If you're generating video and Sora comes out, you're up shit creek. What you do different can make the difference between a successful company or not. I'm bringing this up because it actually happened. This founder called me after Sora release and say, "I'm doing this and I don't know what to do now." But there is a path and we don't have to go into the details in the interest of time. But thinking through where others will go is as important as thinking through where you are going or want to go.. And we spend a lot of time helping companies think through that. Got it. Right? How do you build protectability in 3 years, even though it's so hard? That's the hard part in AI.

That's the hard part in AI.

Is not what to do, but what will get obsoleted by others.

Correct. I mean, it's an exciting time to be a founder, but I actually think that doing the game theory of what everybody else is doing and what the bigger players will do.

Exciting, but really scary.

It is. No, no, it is. And I think, but maybe one thing that I also come back to, which is that if you have a strong team that is nimble, you have the ability to actually essentially in some ways be able to shapeshift and be able to do different things. I think that also, I think that the attitude that you can have within your company especially is that like, listen, next 3 years, are going to be highly turbulent. Even if you think we are winning, it could all be taken away. So we have to kind of stay in some ways paranoid and lean. I think you can actually inculcate that philosophy.

Well, let me take— I'm sure most of you are familiar with things like Cursor. And it'd be disappointing if the nods weren't there. And say, Devin. Yep. When I look at the cursor trajectory, it could build a really successful business. They are building one. But, you know, when you look at cursor and Windsurf and Bolt and Klein and you go down the list and say lots of people coming to the same point, what's differentiated? I don't know where those things will be differentiated in 3 years. Especially when reasoning models get good enough to do most of the code. And if neurosymbolic systems like Symbolica actually succeed, then you have a whole order of magnitude jump in coding ability and planning ability. So thinking through that, you know, we spend a lot of time saying, should we invest in copilots? Like Microsoft Copilot that are helping programmers. And where will the differentiation be in 3 years versus say something like what Devin is doing, which is building a junior intern, which will grow up to be a senior programmer over that time. We prefer the second path to copilots almost always. In fact, we are almost always investing in things that do all the work and for those of you trying to formulate a plan. It doesn't matter what you're doing, do the work as if you're an employee. I tell every healthcare provider, for every physician you have, give them 5 AI intern physicians.

Yeah.

So the senior physician is in supervising role and over time they give them more leeway, just like they would if they hired a fresh MD outta Harvard. So we prefer that model. Because there's places to go. But you also look at a copilot doesn't really understand much beyond the code that they're looking at. When you're building a software programmer, it understands all of the data and context of a corporation. And so you can build much more protectability. I'm just, I'm citing this as an example of how you think through these problems and what's the advantage you sustain in 3 years.

I mean, that doctor use case, by the way, at least from a personal perspective, like just putting in all of your health records into like, you know, O3 or whatever and just having it kind of like asking it questions about your health. It's so much better than even if you have a super fancy rich, you know, like on-call, like, you know, primary care physician. These models are just so good for health, right?

Oh, absolutely.

Yeah. And it's already clearly better. And the second thing I'll say, I think this is interesting, maybe a corollary of what Vinod just said, which is it might actually be a bad thing if you're Cursa right now and you get up to like $100 million in ARR within like whatever, like 12 months. That might actually fuck you a little bit because it might make it harder for you to make the hard decisions as these reasoning models kind of ramp up.

Actually, I think they're doing a pretty good job of pivoting.

Now, but it's hard.

It is hard.

Maybe one last question from me before I open it up. Vinod, how can India, how can the Indian technology ecosystem actually produce more true technology companies as opposed to, well, yeah, I mean, like actually true globally, you know, global tech companies that are building hard technology? What do you think we need to inject into that ecosystem?

The right investors. You know, I like to say my joke about European investors is they turn every great idea into a low-risk, okay idea. And, you know, anybody can make the risk versus reward trade-off. They're just gearing towards low risk and lower outcomes. I like to say most people reduce risk, increase the probability of success, but lose the consequences of success on the other side. And I'd much rather focus on the consequences of success and then say, how do you iterate and pivot and change to increase the probability of success? India question is they have investors who like safety, who like revenue early. Uh, it may not be the right strategy. And I think that's a problem in India is they're looking for a level of safety or financial analysis because they're investors. So we don't allow spreadsheets to be run on IRRs. We've never calculated an IRR in our firm as in technically an investment firm. Because I think it leads to the wrong outcomes.

I mean, I'd say that one of my learnings about starting South Park Commons in Bangalore, and I didn't quite realize this before we actually started our office, is that I don't actually think anybody in India was doing true pre-seed investing. Because even at the pre-seed stage, they were asking for artifacts and progress that don't make any sense. Because at the pre-seed stage, I'm like, are the founders great? Do they have like an idea that they're passionate about? And that's kind of what I'm looking for. But it was a revelation to me that like the risk appetite, even at the earliest stages, was just not there. Yeah.

And it's very hard to do. So it scares people.

It scares people.

Yeah.

Um, okay. We have about 5 or 6 minutes, uh, to get some questions from the audience.

Thanks for the great talk. I'm, I'm a YC alum and every Tuesday we would have those talks from those successful like YC founders, and it would be the same pattern of they had a belief system, they had challenging times, and eventually because they believed in their belief system, like something happened and eventually worked out. It's the same story from the Airbnb. From the thing is like for one successful founder, there's so many that failed like behind. There's like definitely the survival bias. What is a good belief system? Like all those founders, the failed one and the successful one, they all had the belief system. They all thought that their idea was the one. What makes a good belief system and at what point should you question your belief system? Yeah, so I do think there's an answer to that. You start with a belief system of some large change you want to make happen, build a team that's really good at debating all the pros and cons of what's working, what's not working, and evolve your strategy. I think That's not what people count on. Airbnb, if it wasn't for the Democratic National Convention in Philadelphia, I forget what year, 2008 or something, they probably would never have gotten that MVP place from which they could iterate and improve. In most success cases, it's true that they evolved from something. And so building the right culture of debate inside and being able to change your mind. And there's a difference between iterating and whiplashing. And that's a nuanced kind of thing. So it's a balance. If you're too rigid about what you're doing, you're not going to get there. Because something will fail. If you're flexible, you're in a better position. If you're too flexible, you'll end up in the wrong place. I sort of, I have analogies for all this. I'd like to say, if you're crossing a stream with stones in it, you never go straight. You zig and zag a little bit and find the path. On our website is a great article that somebody at INSEAD Business School wrote, Vinod Khosla, uh, after interviewing 400 founders. And, uh, in, in, I'd recommend reading it because it talks about the process of effectual reasoning and how entrepreneurs make decisions at each step. And at each step, they then do enough to gather the resources to do the next step, even though their goal is long-term goal. It's never a linear path. The other analogy I use is if you want to climb Mount Everest and go for Mount Everest, you're never going to make it. If you have to get to base camp first, you can go to— then you have to get to Camp 1, Camp 2, Camp 3. So there's— and none of this is a linear path. It is zigs and zags. And that's a really good analogy for how you get to the next step, gather resources for the step after that. And it's the part of entrepreneurship that's not well understood. And, uh, I'll give you a word of warning: when you have the wrong investors, they focus on revenue. And that's like, instead of going to Camp One, you go to a local maxima that doesn't lead you to Mount Everest, and then you're stuck in this eddy on the side. You reach a local maxima. And so being strategic and thinking through both the long-term and the very tactical. And I like to say, be obstinate about your vision, be flexible about your tactics. If intelligence becomes a commodity, do you think intrinsic motivation diminishes and it loses its value, or would it— become a true differentiator in the future? Are you talking about human motivation? Human motivation to pursue things for purely the joy of learning. But if intelligence is a commodity, then everybody will have access to the same knowledge. Then what really happens to that thought to pursue things? Well, you don't have to be better than the AI to be motivated. If I'm doing music, I set a set of goals for myself. I don't have to be as good as the best singers. It's a personal goal and you move that along. So I do think that'll be improving along the lines you really enjoy, whether it's painting, whether it's gardening, whether it's sports. Human motivation will come from trying to do better than yourself. In the things that you really enjoy. Competition will always be a big thing. Why do we play sports? It's for competition. So there's a lot of human characteristics that won't change.

Before I ask another question, one book that I recommend everyone read that feels as though it is becoming more and more relevant every day is by Iain Banks. It's a culture series. The book is called "The Player of Games." It's talking about a world with sentient drones and AIs. And a lot of what he talks about is that, listen, once we have all of that stuff, we'll probably just invent a bunch of games to play with each other as humans, whether it be in the realm of sports or music or art.

Human ego won't go away, so it'll always be important. This is where influencers started before it became a profession.

It's true.

But let me give you a following example. You can buy a perfectly made glass in a factory or make a somewhat flawed handmade glass. What do we value more? We value the handmade, even if it's not as well made, if it is not as uniform, if it So what we value as human beings will evolve. And I'm pretty optimistic there'll be things for everybody to do.

Pretty optimistic. Absolutely. Well, thank you, Vinod, for coming by. This was awesome. Thank you.

Thank you.

That was another episode of Minus One from the team at South Park Commons. Make sure to subscribe to our show wherever you listen to podcasts and find us on social. At South Park Commons. And thanks to our friends at Atomic Growth for their support in bringing this episode to life.