Y Combinator CEO 谈 AI 时代的创始人心理

We may never achieve a utopia, but it is worthy and worth it to attempt. Everything that's awesome in my life is kind of a cult. We had to go over to the Windows team. They wouldn't reply to our emails. They wouldn't fix our bugs, and we had to go over there with a baseball bat. An org like Microsoft can't.

Yes.

But a startup can.

Yes. Yes.

And every startup must.

Garry, so good to have you, brother. Thank you so much for joining us today.

Thanks for having me.

Amazing, dude. We're going to cover a bunch of the basics, and then we're going to get into the advanced topics to make sure that we match the curiosity and ambition of our audience.

That sounds great.

Does that work?

Yeah.

Awesome. Well, dude, maybe talk to me a bit about the beginning of your journey in 2003. It feels like what's under-discussed is founder psychology, and then Silicon Valley culture, and how the two are connected. What did it feel like when you first got in the game back in 2003, coming out of Stanford?

Yeah. I guess I wanted to work at startups, but 2003 was funny in that Web 1.0 had just ended. The NASDAQ had just crashed, and there were no jobs that I could even find in the Bay Area other than working at Gap IT.

Rough.

The 2 job offers I had were Microsoft for Windows Mobile, which I was pretty excited about, or Expedia.

Okay.

I had student loans, and while I wanted to start a company, I felt like I needed to get a job. There were no jobs in the Bay, so it was hard to believe, actually. That moment was so bleak, right before Web 2.0.

It felt like it might've been the end of tech, huh?

Right. A little bit. Yeah, that was the vibe. At the end of the day, that's the boom, that's the bust. These things sort of happen over and over again.

So at the time, did it feel high-status to work in tech or on startups?

Not at all. It's sort of lost to the sands of time, but I was still excited about tech. I had done a lot of web programming.

Yeah.

One of my regrets, actually, was that I was chasing what was hot instead of what was clearly the thing I knew. Mark Zuckerberg was probably doing Facemash around then.

Yes.

Around 2003. Facebook didn't even exist yet.

Yes.

I was getting out of web programming at exactly the wrong time.

Because it felt like that was the smart thing to do.

It was. Yeah, it was smart, and this is true now. In front of all these people at Startup School, I started this “Don't LARP” meme.

Yeah.

But I think that's embedded in this idea that you should just be earnest.

Yeah.

You shouldn't follow what the crowd is saying. At that moment, people said the web was dead, but that was the perfect time to work on the web, social, and things like that.

Yes.

I didn't do it. I ran off and said, “Oh, well, the next hot thing will be mobile,” so I wanted to work at Windows Mobile. Which was true, but why give up the thing that could've been the next thing?

What I love about what you're saying, though, is that often when we hear “earnest,” we think “naive.”

Mm-hmm.

But actually, what it sounds like you're saying is that earnestness is a result of maturity and courage. Is that right?

Courage is actually the right thing.

Yes.

If you know something, and someone just comes and contradicts you, or even if a blog post or someone on Twitter contradicts you, you're like, “Well, I don't agree with that because I believe my own direct experience.”

Yes.

I know that I'm not a time traveler because if I were a time traveler, I would've sent a message back to myself and said, “Just stay working on web stuff. You literally have 5 years of experience on top of everyone else. You could do an end run around everyone. You could've built any kind of social software.”

Yes.

But instead I was 22, so I was being whipped around by whatever I could read in The Wall Street Journal. The classic question I'm sure you get—the one that all investors get all the time—is, “What's hot, and what should I work on?”

Right.

That's the wrong question to ask.

Yes.

The right question is, “What are you interested in? What do you know uniquely?”

Yes.

That's one of a great many mistakes I made in my career. I feel like I'm a little bit doomed to be the first-principles storyteller of all the mistakes that I made in my career.

Well, I think that it's very inspiring for founders to hear your own journey and your own mistakes, and then not feel so bad about the ones that we've all made along the way.

Yeah. That's right.

I mean, maybe talk about the culture of Silicon Valley today. On one hand, maybe it was despondent 20 years ago. It's crazy to think that was just 23 years ago that you're talking about. Now you think we would all be incredibly elated, yet there's this undercurrent. Maybe it's the LARPing thing that you mentioned. It's being outside of the light cone, the permanent underclass. Where does that come from? Do you think there's any truth in it? Talk through that.

I mean, I get it. There's just more competition than ever. I think maybe it's just that we're spending too much time in distribution. Chris Dixon talks about this all the time, where you're in the idea maze, but the cool parts of the idea maze are the weird fringe parts, actually.

Yeah.

It's always like, “What is a toy?” And that's the pattern in Silicon Valley over and over again that bears repeating, right?

Yes.

The idea that you could have a computer on every desk in every home today is, of course, what's come to pass.

Yeah.

But the people who created that thing were fringe, weirdo punks, actually.

Totally.

It's—

They were outsiders. I mean, Homebrew Computer Club was not where you were getting laid.

Yeah, that's right. You were just following this weird obsession: “Actually, I want my own computer.”

And I think Chris has said this, too: I think a good measure of how interesting the thing is is, can you find the edge? For the most interesting things, you typically can't. Even if you find the edge of the conversation, you can't find the intellectual edge, which I think is a very good sign.

I had this growing up in the '80s. Maybe you had this, too. There was sort of 1 way to be cool in grade school—

Mm-hmm.

—and probably in high school. Maybe when you got to college, you kind of found your people. Now I think one of the miracles of the Internet-native generation is that people find their people much sooner. You find your people on Reddit, on X or Insta, or whatever your network is.

Oh, yeah. That's the coolest thing today. There's probably a subreddit, probably a part of X, some part of Twitter that's just your set of super-weird people who are really, really into that thing.

Yeah.

You can find your tribe more or less instantly.

Yeah. Yeah, can you talk a bit about the culture of Silicon Valley? It felt like 10 years ago it was lower-status to work on something like Palantir, either because people were nervous about doing things that were too much in the national interest, or because it was dismissed as a services business. Now it's clearly one of the most important companies in the world, with one of the most compelling CEOs. You were early there. What did it feel like? Why did you do it?

Yeah, I mean, I have a worse story than that. I have another mistake story, where Joe Lonsdale and Stefan Cohen were my fraternity brothers at Stanford, Phi Psi.

Wow, okay.

I had a job, and they were interns at Peter Thiel's hedge fund.

Peter has this thing where, when you start a thing, you're supposed to write down on a piece of paper all the smartest people you need to go and hire and get to join you.

And so they both said, “Oh, let’s go get Garry. He’s up in Seattle.” They flew me down—

Yeah.

—to have dinner with him. Peter said, “I’m so sure this is the right thing for you. Here’s a check for $70,000.” That is how much I made at Microsoft at the time.

Yeah.

And I said, “Thank you very much, Mr. Thiel, but I might get promoted to level 60 this year.”

Oh, no.

Which I did. But I think it’s—

About what cost?

—a $2 billion–$4 billion mistake at this point.

Oh, no.

I think about it all the time. Basically, I did the same thing again. It was the same thing that got me to go work on Windows Mobile instead of—

Yeah.

—just staying on the thing that I knew and loved.

It happened again. I was working backward from the map instead of looking down at the territory and saying, “These are the smartest people I know.” Joe and Stephen were among the smartest people I’d ever met.

Yeah.

And if not them, who was I going to be a co-founder with?

Yeah.

The universe was saying, “This is the thing you need to do.” But all I cared about was what was cool, what was hot, and what a really good investor would say. The thing is, they’re going off a map.

Yeah.

You can see, you can look down, you can look around, and you can have first-principles knowledge of what’s going on.

Yeah.

What they were doing was going to governments and 3-letter agencies.

Yeah.

And it was clear that the things that really good computer scientists in Palo Alto could do and create—

Yes.

—D.C. did not have access to. And it’s funny how much the thing that Peter, Joe, Stephen, and later Alex Karp identified—

Yeah.

—runs through almost everything that we do—

Yeah.

—in Silicon Valley.

Yes.

The future is already here; it’s not evenly distributed.

Yes.

And that’s where first-principles knowledge, or just viewing the world as a scientist, is actually very useful.

Yes.

What a scientist does is, sure, there are published papers, but you go and examine all the truths.

Yes.

What you find is that once you go down into the experiment, or go down and talk to the first-party people who are experiencing a thing—

Yes.

—then you learn secret knowledge; then you have gnosis. From Palantir, I definitely learned that everything that’s awesome in my life is kind of a cult—

Yeah, yeah.

—that starts with some sort of truth or belief that flies in the face of an orthodoxy, which is very punk, actually.

Yes. Totally, dude. That’s very Rick Rubin—he would approve.

Yeah.

Well, actually, what I love about what you’re saying is that there’s a beautiful truth in the fact that you’ve talked through some of these maybe tragic misses, but the lessons are courage; the lessons are earnestness, right? The lesson is to follow your instincts when they’re pulling you to work with a set of people. So maybe talk a little bit about how Silicon Valley’s magic is that it takes outsiders and turns them into insiders. We talk a lot about that, and I feel like YC is the ultimate distillation of it. You go from being 2 founders, perhaps from a foreign country, to 2 people with YC on their LinkedIn, and everybody now knows what that means.

Totally.

So talk about how that happens. Is it a process? Is it a mindset? And how has it changed over the years?

I have to credit Paul Graham, Jessica Livingston, and the original founders, including Trevor Blackwell and Robert Morris. What Paul and Jessica figured out was that there are lots and lots of really smart people who do not get access to Silicon Valley. In the past, I think you had to go to the right schools, know the right people, and if you didn’t, you’d come to Silicon Valley and try to get invited to the right parties. There was a social network that you had to crack. At YC, we have a pejorative for that whole scene: we call them scenesters, actually.

Okay, okay.

It’s a nod to what you used to have to do. Some people still believe that’s how you do it.

Mm-hmm.

All they did was say, “You know what? Here’s a website. Here are 12 questions.” Later, we added a 1-minute video.

Yeah.

And people on the other end who are builders are going to try to give you a fair shake. Today we have 16 equal partners, including myself. All of us went through YC and had successful outcomes, and I think there’s something valuable to that: “Welcome to Silicon Valley.” We just celebrated 7,000 people coming to San Francisco, the majority of them coming for the 1st time. It’s our Startup School every single year.

So incredible.

We call it a birthright, but for tech.

It looked like a Roman Colosseum, dude.

Yeah.

It was amazing.

Yeah. I think that’s the coolest thing about Silicon Valley.

Yeah.

It sort of doesn’t matter where you’re from or who you are. It actually matters what your ideas are, and whether you can execute. The only other thing that I’d say YC people get that a lot of people don’t get is a community of people who—

Right.

—you can be super real with.

Right.

If you go to a TechCrunch event, you look to the left, you look to the right.

Yeah.

Are you really going to be able to tell them anything real? You’re going to hold your cards pretty close, and you’re going to say, “How are you doing?” They’re going to say, “Killing it.”

Yeah.

And then they’re going to ask back, and you’ll say, “Oh, I’m also killing it.” Honestly, founders really need that. It’s very solitary—

Yes.

—on the day that you lose your best customer, on the day that your best engineer quits, who are you going to call? If you have no one, I mean—

The day your co-founder loses hope—

—a lot of people have no one.

—and you’re not allowed to.

Right.

That’s a tough day, you know?

Yeah.

Talk about the historical upside surprises of founder types, business types, and the downside surprises. What have you learned through YC?

I think we’re in the middle of a big transformation now.

Okay.

For the longest time, we really, really believed that you needed co-founders. Everything else being equal, having co-founders is net really, really good.

Totally.

I think some of this is the beginning of a cult. A cult where there’s only 1 person, and that person can’t convince anyone else to join them, is—

Worse than—

—worse than a cult.

Yes.

It’s just a crazy person, actually.

Right.

I think this is an old Derek Sivers TED Talk, right?

Okay, okay.

1 person dancing in a field is a crazy person.

Yes.

But it’s really the 2nd person—who gets up—

That’s right.

—and joins that person. Then what you find is—suddenly, 10 people are—

It’s a party.

—all dancing. Then it’s a party. Yeah.

Right.

That’s literally the TED Talk.

Love it.

And I think there’s something to that. But what’s happening right now with vibe coding and agentic coding—

Yeah.

—literally any given person could be 400 of that person. From 2 years ago, even 9 months ago, you can be a multiple of yourself. And so—

Isn’t it amazing?

Yeah. And so—

Do you think founders are more ambitious than they’ve ever been?

They should be.

Yes.

If anything, that’s the number one thing that people should do—

Yes.

—is realize this: don’t chase your success or my success.

Yes.

That was when we were coming up. The game has changed.

Yeah, that’s right.

You can’t just do a SaaS thing. I was just doing an analysis on pure SaaS, and I was like, “Oh, shoot.” A pure per-seat SaaS thing—it’s not totally clear that it will exist in another 5 or 10 years.

It's perfectly fine as a wedge, but—

Yes.

If you're doing that now in 2026, you better hope that that's a leap to some sort of moat around data or network effects or something else, right?

Yes.

And if we time-traveled to even 2 years ago, SaaS was God.

Yes.

You know? The 10–20x next-12-month revenue multiple was—

Yes.

—an iron law.

Yes.

It's hard to believe. It was like, "Oh, yeah, no, that's not true at all now."

Well, all this is a related thing that I love: code is no longer precious.

Mm-hmm.

My God.

That's the best.

The things that we had to go through to get something shipped—

Yeah.

—and that's why you had to have this sense that what you were making was important with a capital I.

Oh, yeah.

And that actually flies against the kind of beautiful truth we were just talking about, because often earnestness isn't obviously important with a capital I. So you pursue things that seem important, when most of the most important things actually start out seeming unimportant.

Oh, yeah.

I know you've been all over this with GStack and GBrain. When you started out to build those things, was the intention to contribute to the community or just to explore your interests?

I was just messing around. I think Boris Cherny came on the Lightcone podcast. A bunch of our mutual friends basically just started coding again, and I was like, "Oh, something's happening. I'd better start."

Yes.

Initially, it was just messing around. What would it be like to make an Office Hours?

Yeah.

We had just started recording all of our office hours, actually, and then I had a few months of data. There were very clear things that the YC partners used as rhetorical devices or ways to draw out a step function in your ambition.

Mm-hmm.

What's funny is I have this Markdown file that's basically the perfect script on how to advise founders to increase their ambition.

Yeah.

I actually had to go into Claude, and I said, "Reduce this by 90%. Reduce the strength and intensity of this prompt by 90%."

Yeah.

Then I open-sourced that, because some of it is, "You gotta come and—"

That's like—

"Do YC—"

Can my team get the 100%?

Yeah.

Of course. Of course.

It's like, well, we'll take—

I can't give away the—

Your Fable 5.

Okay. Thank you.

Yeah. We can't give away the crown jewels for free.

Yeah.

I'm like—

Right.

This is just a taste. The first one's free.

Well, dude, Kevin caught this. I think it was in GStack or GBrain—

Oh, yeah.

—where, deep in the open-source code, you had said, "Please apply to YC."

Yeah, but only if you're a top 10% user of Claude Code.

Incredible.

Based on whether you're thoughtful about your users, your product, and your design. Not everyone gets that message, actually.

Yes. Well, dude, I love that you're doing it, and I love that you're doing it so publicly, because I think this comes back to the earnestness point, which is that the whole thing about building intuition at this moment is just using the technology. I think there's this feeling of, "Oh, how can I possibly catch up?" There's so much seriousness around it. It's like, no, just use every model and build completely trivial things for no other reason than to have a chassis with which to learn the new technology.

The most exciting thing is that you can just have ideas, and in the past it was like, "Oh, well, I had to be the PM."

Yeah.

Let me write the spec.

Yeah.

Let me make sure I have one or two engineers. How are we going to do QA? There's just all this stuff that you used to have to worry about.

Yeah.

And now it's like, no, `/qa`—

Yeah.

It's done.

Yes.

But we tested it. It's good. We have unit tests, 90% test coverage.

Totally. It's like arranging your Zen rock garden. It's just fun to do it, and the output doesn't matter that much, which is why everybody talks about coding agents but very few people are showing their work. A lot of it is low-stakes work, and I think that's beautiful and okay.

No, I think it's crazy to what degree suddenly agency and taste are off-the-charts important now.

Right.

The hard part is helping people realize that just by the act of doing it and exercising it, that's how you get better. I don't actually think that those things are innate or inborn. Some people have more of it coming out of the womb, per se, but it's a practice.

Sure.

Well, let me challenge you on this, because Nikita said this a couple—maybe a year ago—and I thought it was interesting. He said, "Hey, we were told that the highest form of art in business was film and software. Now anyone can make films and anybody can make software, yet the same 20% who were originally doing it are still the ones who are doing it." Is there kind of this—is agency something that you can embrace, or are there just people who are naturally high-agency and many who are not?

If I look at my own career, I would easily say that I'm a late bloomer.

Okay.

I had to make these ridiculous mistakes, like saying no to Palantir or leaving web programming to learn. And so I feel like my life is somewhat a testament to this: I can make mistakes, introspect, reason about them, and make different choices.

Amazing.

That's the hope, anyway.

Yeah, that is the hope.

Yeah.

Talk a little bit about loops. I think there are coding loops, which are well discussed, and now the graphs thing, which is both funny and real, but it feels like the most interesting loops are business loops. How can you actually have a loop where the output is changing the way the business itself operates? I know Tom gave this great talk about how to organize your company in that way.

At GBrain, yeah.

Talk about how YC is organizing around loops, and how you're advising your companies to do it.

Yeah, absolutely. I mess around with OpenClaw. I created GBrain, which is kind of like SOTA retrieval. Earlier this year, I created garyslist.org, which is basically an agentic newsroom. Then I had to create really good RAG around personal deep research just to be able to put together some of those articles.

All I did was have OpenClaw look at that repo and extract the agentic RAG part, especially around after-the-fact graph generation. Then it just ended up being SOTA even against Memory Palace and a bunch of other open source.

What I realized is that just allows the agent to know which million tokens to keep in its head for any given task. I think there are a couple of things at work right now. The frontier model labs that are letting you use ChatGPT or Claude, I think, are still pretty constrained on cost and on the amount of compute they want to give you.

Mm-hmm.

If you really want to token-max, you actually have to use something like Hermes Agent or OpenClaw.

Mm-hmm.

Then you have to tune it all the way up. You're just like, "Give me—let me load 1,000,000 tokens or 800,000 tokens into any given request," and have that be in your sol.md.

Yeah.

But when you do that, I think you basically get to live in 2028. It costs, I don't know, $50,000 or $100,000 a year to use the agents at full strength—full 150 IQ—

For every request, for every interaction.

Like, for every request.

Mm-hmm.

But you'll feel it more or less immediately.

Yeah.

It just costs a crazy amount. But for a CEO or a founder, it actually makes a lot of sense to do that, and you have to give yourself permission to token-max in that way. And then you get to live in 2028 today.

Amazing.

But I think, going back to what founders can do, if you live that way, not only do you want to do that task, you want to skillify it.

Mm-hmm.

You do some feat of strength, and then you turn it into a Markdown file, plus code, plus tests, that can be reused and put into a cron job.

Yes.

I think we're just seeing, across the board, a Markdown file is an employee.

Yeah.

And it's an employee that will do the job perfectly every single time.

Yeah.

And it'll do it as many times as you want.

Yeah.

At that point, all you're doing is taking any business process that you need in your company, and you just do it once perfectly. The first time you try to do it, it's going to be bad.

Expensive.

Yeah.

Yes, many iterations.

Yeah. But then at the end, as long as you can tell very quickly and in very simple ways, “Hey, agent, fix this, fix that”—this was wrong—the actual trace and history of that agent working with you will actually turn that into a skill file that's perfect.

Right.

And any time it screws up in a future case, it's just a bug fix, and then it's there forever.

And this loop you would apply to sales, to marketing, to customer support, to everything.

Yeah, to everything.

Really interesting. I mean—

GStack was basically that for engineering, right?

Yeah.

There’s a QA loop. I had to build something that works with browsers, because that’s what I found. Early with GStack, I found myself automating how I thought about PM.

Mm-hmm.

And then I had an engineering skill that was like an engineering manager, making sure that there was unit-test and end-to-end-test coverage.

Yep.

But in the end, I found myself automating all the other things so much that all I was doing was black-box testing to verify. And it was like, “Oh, no, no, no. Okay, even that part.” Building these agentic systems today, I think all you're doing is seeing where the bottlenecks are.

Yeah.

And then almost every single time, it's, “How do I just instruct the agent to create a piece of software or a Markdown file that blows away that roadblock?”

Right.

And then when you take a step back and you've done that for the whole task, there are companies now that you can take from $0 to $15M ARR in about 4 months with—

Crazy.

—2 or 3 people—

Yes.

—and a few hundred skill files.

It's amazing. So there's this recursive self-improvement. There's this hill climbing that happens. At what point does it start to fall over? Do you need error correction? Do the errors compound? Does it get constrained by exception handling and human relationships? Otherwise, these companies should just be able to grow and run forever.

Yeah. I mean, once you get a lot of Markdown files—any time you have a large amount of data—that's where provenance matters. That's where conflict management matters, too. When you have 2 facts, if you have provenance, then you're like, “Well, this is the more recent one. This is the one that wins.” You need a cron job that's sweeping through that and making sure that things are accurate and from the right places. But these are things that, as you use the systems, you just know you need to build.

Yeah.

Anyone who's experienced this—I mean, this is actually one of the interesting things that I think is an advantage for founders in their 30s and 40s now.

Yeah.

I think there are going to be as many Patrick Collisons as ever.

Yeah.

But one megatrend that we're seeing is 35-, 40-, and 45-year-old founders who've been around the block.

Yeah, yeah, yeah.

Built a lot of entries.

Don't count out the OGs.

I think Peter Steinberger is a perfect example of that—

Yes.

He’s been around the block.

Yes.

He knows what to build.

Yes.

And then if you take that person, suddenly there are 400 of those people. You can outperform an entire department of any Mag 7.

I also think that if you've had some experience in the enterprise, there are these dynamics that you can seek to mirror and explore. One of the things I'm really interested in is conflict. Most people at an enterprise would say conflict is a bad thing, but I actually think that there's this productive, constructive conflict that, at its best, is the search for truth.

Yeah.

These systems allow us to do things like explore conflicting approaches and totally unbundle it from the human emotion.

Oh, yeah.

And I think that's actually very important, because this is why decentralization in an organization or an economy works well—

Oh, yeah.

Every idea gets tried.

Yeah. Do you know Pedro from Brex? He's the most aggressive person I've met who does this with his exec team.

No.

Which is cool. He actually was one of the people who really got me into OpenClaw.

Okay.

He came in and had lunch with us at YC. He walked us through it. The vibe around OpenClaw was that it was too unsafe and dangerous, and that it would delete your whole computer and stuff like that.

Oh.

What he ended up doing was building an entire open-source layer called Crab Trap, which he has actually open-sourced. It's an agent that watches all the network traffic and everything that OpenClaw does.

Nice.

That's how he squares the circle. Brex is one of the most regulated fintechs—

Of course, yeah.

Of course.

Yeah.

The upshot of what he got is that he has his agent look at the meeting transcripts from all of his direct reports, and he can see 2 levels down.

Yeah.

The thing that really jumped out at me is that he can use meetings that he's not in—

Yeah.

To figure out what's broken, what's in conflict, and who's in conflict with whom.

Yes.

Sometimes he can show up at a meeting with perfect context from the last 3 weeks of that one compliance team.

Yes.

He can walk in and say, “Actually, you're right. We're doing it your way,” and then he can leave.

That's so badass.

Yeah, and can you imagine? That is actually one of the defining reasons why startups or businesses fail. It's actually a management challenge. The business becomes too big to fit in one person's head.

Yes.

This is the most profound form of it: You have agents, memory, and retrieval.

Yes.

You zap the logs more. The logs give you raw intelligence about what's happening, and it can actually introspect not just on the opinion of someone 2 levels down in your organization—

Yes.

It can actually look at the ground truth from the data.

Yes.

Right? And that's clairvoyance, right? Once you have that in your brain, every business in the world is available for you to take their business.

Yes.

Right?

Yes, yes.

The world is basically built on things that are built around human beings, and human beings can only keep 7 ± 2 things in their head—

At any given time.

Yes.

And guess what? You, plus an agent, can keep basically 3 Harry Potter books in your head.

More to come.

Yeah.

Yes, yes.

And more to come, right?

Yes, yes.

There are so many crypto-isms that apply 10 times more to AI. This is the worst it will ever be, actually.

Yes.

And so the whole world—all business, all the things people rely on, especially government—is basically still built on a world predicated on human beings, limited human beings who can only keep 7 plus or minus 2 things in their head.

Yes.

And it’s not ready for the world where agents can think at the speed of light. You have unlimited amounts of it—unlimited amounts of data. My hope is that products are going to be 10 or 100 times better. They should be 1,000 times better.

Yes.

Services should be 1,000 times better.

Absolutely.

Right?

Yeah.

The invisible hand of capitalism, properly done, isn’t about lowering the cost of something. It should actually be creating these sort of magical things. I was on The Wall Street Journal podcast recently.

Yeah.

The reporter kept trying to gotcha me. He kept saying, “Are you guys in tech trying to make a utopia?” And I was like, “This is a trap.” I would not go so far as to say that. But as I went through a bunch of examples from the startups that you and I have worked on and funded and things like that, I was like, “Actually, maybe.”

Yeah.

Right?

Yeah.

We may never achieve a utopia, but it is worthy and worth attempting.

Yes.

I would not say that Silicon Valley or tech will achieve a utopia. I will very earnestly say that we are trying to bring about a better world.

Maybe another way to say this is just: this is the least happy you’ll ever be. Because I think one of the other implications of what you’re describing is just the ergonomics of the white-collar job today. You’ve worked at big companies. You think of just the kind of morass that you had—

Yeah.

—to get through.

It’s torture.

You think of how lossy the communication is. The entire management science—the managerial class—maybe that was a 100-year distraction from what was founder mode, in the Industrial Revolution, all the way through Henry Ford, all the way through some of the early Silicon Valley founders, and then everybody just hit this tidal wave of management.

Oh, yeah.

Now we have this new model that can lead to dramatic productivity improvements: 3 days a week of really intentional work, maybe 1 day.

Yeah.

And the other days, you’re touching grass. I think the time that you do spend is just so much less burdened by the stuff that you always felt like you had to wade through.

Yeah, that’s right. If you look at the typical day-to-day of the Magnificent Seven, it’s insane.

Yes.

I remember when I was 22, working on Windows Mobile. We were working on these crazy devices that were years before the iPhone. We needed integration, for instance, from the Windows team, which was on the other side of the highway in Redmond.

Right.

My PM mentor and I had to go over to the Windows team, and they wouldn’t reply to our emails. They wouldn’t fix our bugs. They wouldn’t even mark them as “Won’t fix.” They just ignored us.

Yeah.

And we had to go over there with a baseball bat.

Yeah.

I don’t know why he brought one. It wasn’t like he was—

Are you kidding?

—going to rough anyone up. It was just this vague threat or something.

Incredible.

It was actually hilarious, and I think about that all the time. What kind of bureaucratic hell is it where two perfectly smart people who are making good tech money are put into a situation where there’s a fiefdom over here?

Yeah.

Literally, there are 7 levels between us and that person.

Yes.

And this person doesn’t even know that we spent 4 hours trying to get a P3 bug fixed that was a blocker for 1,000 of our users—or something.

Yes.

And how much humanity is just wasted?

Yeah.

How many 2:00 a.m. anxious wake-ups were there because somebody wanted not to have a hard conversation? It’s a real waste of the way we should be spending our time and working.

Yeah, and that’s a profound version of the 7 plus or minus 2. I don’t think that the product unit manager or whatever the heck their title was—

Yes.

—I don’t think that they wanted us to waste our time like that.

Yes, yes.

I think it’s just that your day—you just don’t have enough hours in the day. We’re incredibly limited human beings, but—

Yes.

Yeah, it’s funny. I think about my time at Microsoft, and I remember getting an iPod and doing competitive research for Windows Mobile versus the iPod.

Yeah.

Today at Microsoft, you could imagine a Markdown file and a skill—

Yeah.

—that is run on a cron.

Yes.

How are all the people working on this team? Are they blocked? Who are they blocked by?

Yes.

What are the external dependencies?

Yes.

Suddenly, there is a graph.

Yes.

There’s a loop.

Yes.

And then suddenly, the products can actually just move faster. You could release something in a day or 2 that would have taken 6 months.

That’s incredible.

Right?

And if you could multiply that out across an organization like Microsoft, the thing is, an organization like Microsoft probably can’t—

Yes.

—but a startup can.

Yes.

And every—

Yes.

—startup must.

Yes. They must be organized this way. I also love what you said. It’s almost like the mid-level of bureaucracy should be agents, because I actually think that being in charge is somewhat overstated.

Mm-hmm.

I’d love to actually have at least the mid-level organize the work, resolve the conflicts that it can, somewhat programmatically—

Yeah.

—probabilistically, programmatically.

Yeah.

Of course, the executive team maybe sets direction, the individual decides on execution, but just the whole organization layer, which a lot of product managers historically did, should not be human.

Yeah. The great thing is, it doesn’t have to be—

Right.

Do you know how Venkatesh Rao has this thing where it’s like you need to be above the API line?

Right.

If you’re an Uber driver, you’re below the API line.

Yes.

You should be fighting to be above the API line.

Yes.

I actually think AI erases the API line.

Yeah.

One of the funniest things I saw recently is that there’s an API company that we funded called Clawvisor, and they released a bug-report endpoint that isn’t for humans to use. It’s for agents that use their software. There’s an LLM that, if there’s a bug in the software, the bug report goes in, and then it replies in real time: “Actually, that’s a good bug. We filed it. We’re going to fix it next week, but also, try this workaround.”

Wow.

And I can’t get that out of my head, because you could imagine that—that could be how companies work.

Yeah.

Instead of being below the API line and having to do whatever, I don’t know, the API tells you to do, this can be a back-and-forth, actually.

Yes.

And I always think about why cars in the 1980s got so good in Japan…

Mm-hmm.

It was actually a management technique. They realized the Toyota Production System—

That’s right.

—it meant that if you are the person doing the work on the line, you should be allowed to change the line, because you are the person with the most context—

Yes.

—to improve your own working situation, and that’s why American cars really started sucking and why Japanese cars became absolutely incredible.

Yes.

It’s a very subtle, simple way that actually empowers the worker.

Yes.

And so there’s this mix that I think will happen.

Which is so interesting, because what’s traditionally seen as a hierarchical society came up with this very bottom-up—

Beautiful.

—way of working, which led to arguably a 40-year generational run because of process power.

Yeah, which is beautiful.

As you think about our evolution toward this vision, how many of these problems do you think are intelligence-bound—meaning we need smarter models—versus there being other attributes of models that will unlock success?

Honestly, it’s all human. Humans are the ones who are going to slow this down, and I’m realizing this might be a white pill.

Okay.

If you look at the tenor of what’s going on broadly in the discourse about AI, it’s all doom and gloom: all white-collar jobs are going to go away, and there’s going to be a permanent underclass.

Yes.

Funny enough, what I think is that all of the bureaucracy and slowness, the 7 ± 2 of every company, every institution, and every organization—all the middle managers in the world—

Yeah.

—that is actually the white pill.

100%.

These things are going to be slower than you think.

Yes.

We had 7,000 people, and most of them were 18 to 22 years old. I looked in their faces, and they were all AI-native. They became adults in the time of ChatGPT.

Yes.

For us, we now run shit, which is nice.

That’s great, right?

Imagine us 20 years ago. We were the ones who were native to the web.

Yes.

We were the ones who were native to mobile.

Yes, yes.

Now we run shit, so all of society runs on the things we expected when we were 18 to 22.

Yeah.

The white pill to me is that it’s going to be 20 years, and that’s not a bad thing.

Yeah.

That’s actually a good thing.

Yes.

Society is way slower than you think. Government is way slower than you think.

Yeah.

Every company in the world is way slower than you think.

Yeah.

There are real moats to these things. There are structural reasons why Microsoft isn’t going anywhere.

Yes.

I guess this is a little antithetical to what most people in Silicon Valley would say. I want a startup to replace all of them.

Yeah.

The reality is, they won’t.

They won’t. I love it.

Okay, I have 2 more questions for you. One is on trying to envision the near future. What do you think the next computer looks like? What does a new computer look like?

Oh, man. In the short term, it’ll probably keep looking the way it does, but it seems inconceivable that it’ll stay in that form factor. It’s pretty clearly going to be voice.

Yes.

I think there’s a lot around memory.

Yes.

It’s going to take a while. We’re at this moment where everyone’s sort of happy with ChatGPT, Claude Cowork, or whatever it is.

Yeah.

But I don’t think they will be in the long term. I think you’re going to want a lot more context, and you’re going to want “All Watched Over by Machines of Loving Grace.” You’re going to want this thing that is benevolent, knows about your hopes, fears, and desires, and is relentlessly searching to help you in that way. That’s actually going to be really exciting.

The path through that will probably be computer use, memory, diarization, ingestion. I think there’s going to be a lot there.

The cool thing is, it’s actually going to take longer. I don’t think people have even been talking about this now. My guess is that next year, 2027, will be the harness wars.

Yes.

You know? It’s really a cost thing. It’s just too expensive right now.

Yes.

But it won’t be. There will be a war for a billion consumers right when today’s frontier-model compute will be maybe $50 or $100, in 2 or 3 years.

Yeah, yeah.

That will be when the browser wars are back on.

Yes, yes. We’re already seeing it: the price per token of the frontier is going to infinity. Mythos [?] is infinity dollars a token. You can’t have it. And the price per token of the model that was frontier a week ago is collapsing to zero.

Yeah.

To me, that is a huge white pill for a lot of reasons. One of the most important is that we need more consumer AI software. In a world where most consumer products are free to try, you just can’t do that when you have really high costs of incremental distribution.

The magic of software is zero-cost incremental distribution. We have these magical primitives. We have computers that are emotive and have emotional texture for the first time in 40 or 50 years.

Oh, yeah.

What a tragedy that we haven’t built more ambitious consumer products yet.

Yeah.

I’m really excited for the changing economics of models to unlock these new categories.

Yeah. We’re so early. Becoming a convert to OpenClaw made me very convinced that it was going to happen right away.

Yes.

At this point, I’m like, that’s not correct at all.

Well, you should try the new ChatGPT desktop app. It actually has a voice mode, and it’s an Operator or a concierge that can see all your threads. You can access it from your phone.

Oh, cool.

It’s very OpenClaw-ish, and I’m sure Peter had his hands all over it. To me, it’s the closest thing we’ve seen yet to a true consumer assistant.

Oh, I can see that.

It’s awesome.

Yeah, yeah. ChatGPT always had basically the best memory, so I’ve got to try it again.

They’ve done a great job.

Yeah.

Okay, last question. We talked a lot about utopian visions for the world and problems that are maybe intelligence-bound. You’ve also put a lot of energy into local politics and doing things I think are very prosocial, but it seems like more of those problems are human-coordination-bound.

Oh, definitely.

How do you get energy to do that work? Is there a white pill for that aspect of our society? Maybe talk a bit about that.

The white pill is just believing that America still has rule of law, and that people in government are trying to help and are somewhat earnest. It’s not perfect.

Mm-hmm.

The white pill for me recently was just doing jury duty in San Francisco and seeing this orderly process involving a real cross-section of San Francisco. One thing I realized was that we’re actually a big part of the city. About 10% to 20% of the people are directly working in tech.

Yeah.

The judge was really good. It was a horrible case, but government is not here to ruin people.

Yeah.

And then, at that point, we can vote.

Yes.

We actually have a say. I think everything is downstream of media at that point. The San Francisco moment that got me very engaged was realizing that Asian Americans are 25% of the voting populace and about 30% of the population, but during COVID, we had a district attorney who was turning a blind eye to some of the most heinous crimes against our elders.

Mm.

At the same time, we had a school board that was openly hostile toward Asian American kids who just wanted to learn algebra in public middle school.

Which hit me because I was that public-school kid in Fremont in the East Bay. If I couldn’t have taken algebra in middle school, I couldn’t have taken calculus by senior year, and I wouldn’t have been able to study engineering at Stanford. I wouldn’t be here doing any of the things I get to do today. And I’m like, “Oh, okay. Now they’ve made it personal.”

Yeah.

And let me understand. It turned out to be all the same people: the same people who want to destroy all data centers, the same people who don’t want to build any housing anywhere in California or the world.

Yeah.

They’re NIMBYs who want to virtue signal, who have these ideologies that don’t square with mine. As a builder, you’re like, “Well, words have meaning. Supply and demand exist, markets exist, incentives exist.” And then I think that when people like us don’t speak up, when we don’t pay attention and hold our media accountable—not just the government—then things get very deranged.

Yes.

And that’s what San Francisco was, I think, around COVID. Then, just by sheer organization, the Chinese of the West Side organized around algebra and the deaths of our elders, going to farmers’ markets and literally telling each other, “They’re killing us. They’re killing us,” right?

Yeah.

ABC News, literally the mainstream-media TV channel of San Francisco and the Bay Area, would not cover anti-Chinese and anti-Asian crime.

Wow.

And it’s unconscionable, right?

Yeah.

Our institutions failed us. The great thing is, when the system becomes so odious, you’ve got to put your body on it.

Mm-hmm.

I think that was Mario Savio.

Mm-hmm.

It’s ironic almost to say that, but on the other hand, that’s what I believe. They can’t do this to us. We’re going to stop it, even at personal peril, even at reputational cost. We should be willing to take that cost—

Yes.

—because that’s our covenant with our fellow man and our fellow citizen, you know? And if we don’t do it, who’s going to do it? Nobody does it. I’m super proud of what we’re doing in San Francisco because we have beaten back that ideological set of people who created this mess.

It’s not perfect in San Francisco, but it’s getting better—

It’s amazing right now. I mean, I think I saw that Lurie might be the most popular mayor or politician in the entire country. Things have gotten so much better. Everybody sees it, feels it in the vibes. Some people say, “I wonder why,” but I think for those of us who know, we see that your stewardship—not just of the industry, the startup community, and technology, but also of the community—is one of the big catalysts. So thank you for doing that work.

Thank you so much.

Yeah.

I think San Francisco should be the beacon for pushing this back in every blue city—

Yes.

—in America. That’s what garysviss.org is basically. How do we take the lessons that we learned here? Los Angeles needs this. New York needs this. Minneapolis needs this. Seattle needs this.

Totally. And also, to connect the 2—the culture, society, and technology—let’s find very local ways for the ambition that this technology unlocks to show up. I was in London last week, and we were talking a little about what’s possible. It’s like, hey, you don’t have to have an American vision for how AI impacts your society. If you want the NHS to be incredible, then you can do that with this technology. So the opportunity for ambition to be very localized in our communities, I think, is something really interesting.

Yeah. Act local. Take care of the people right next to you.

Yeah.

I think it’s funny: so many people get very, very obsessed with the national level. I mean, we want to look at the national level generally right now.

Yes.

It’s like, you know what? I care about San Francisco. I care about housing. I care about crime. I care about treatment and recovery.

Yeah.

And if we do that, we can have the society we want to live in. Then, if we fix local, state and national will fix themselves, because on a 5- or 10-year basis, the people who win up there will come up through a system that’s functional and about saying no to things that don’t make sense—

Absolutely.

—about saying yes to the things that are awesome for the people.

I love it. All right, Garry. First of many. Thanks, brother.

Yeah, man.

Thanks, bro.

This was great.

Cool.

Yeah.