所有提供商如何应对 Anthropic 依赖风险及 Figma IPO 拆解：最终定价会落在哪里？

This idea that oh woe is me I can't raise a third fund I've never returned any capital tough luck right almost everything about a big fund is good for the entrepreneur anti-portfolio regret is the psychological price you have to pay for being in the game because it's literally the emotional tax you pay for being in good deal flow the market for consensus is fully priced in and fully discovered. Guys, I'm so excited for this, as always. It's my favorite conversation of the week. Rory, we have a celebrity in our midst. I don't know if you saw social media over the weekend.

Yes, we do.

But we have Jason here.

Viral, baby.

Jason, I would love to start with you and your experience over the weekend vibe-coding. What did you learn from that experience when we did this last week?

I've learned so much in 1 week. It's crazy. It's the biggest fire. I think your investment in Lovable is even better than I realized a week ago, because these apps—you can't stop. It's a tsunami. Vibe-coding has just started.

We're only 6 months into non-developer vibe-coding, and we're less than a year into it for developers. The things you can do on these platforms are things you could never do before, and that's where you make money in venture. That's where you get the big ones: when you can do something you couldn't do before. I'm in love.

The flip side is that I'm embarrassed, Harry, that we've been talking about Cursor, Windsurf, Claude, and all these big numbers and rounds, but I never understood the general topic of safety—what an agent can do, should do, and shouldn't do—until all of this. That was the meta-learning.

For anyone who missed it, basically you had Replit Agent basically kill your database. Can you explain what happened so people have the context?

I was just using a vibe-coding platform that is used by many, and I did not understand that the agent helped me build the website and the platform, but could also change any line of code, even when it was out in the wild, even when it was in the real world.

For folks who haven't done it, they'll need the context. We have preview, staging, and production servers. It's been done since the dawn of web software. You have preview, which is what we work on in the office. Then, when you're ready to go out into the real world, you put it up on staging. You carefully test it; it simulates the real world, but you're not exposing it to customers or customer data. When you're comfortable on staging, you flip it into the real world: production. That's locked the hell down. It can't be touched.

For these vibe apps to roll at the pace they do—which is, Harry, I've been addicted for real, not as a joke—you saw me. I missed board meetings when I was vibe-coding. Even last night, I got a DM from the CEO of Gorgias saying, "Are you coming to the board meeting this week?" I'm like, "There's a board meeting? This is not a joke." I forgot. I didn't read my emails. I was addicted. I love vibe-coding. I love it. I love these apps.

One of the reasons they're so fast and agile is that everything shares the same database. It all shares the same code. That's really fun when you start a project because it works at light speed. It's magic. Go to Lovable, or any of these apps, and type in, "I want to build an app for VCs that do weekly podcasts and share data. I want to see who's in what deal. I want you to research it. I want you to force-rank them each week for fun, and then email everyone and tell them who's best." You can build that in 30 minutes. It's so cool.

I didn't realize, because I was taught from day 1 that you have these distinctions—preview, staging, and production—and it's all one. What I also didn't realize is that this is both a feature and a bug. Maybe other AIs work differently, but Claude, by nature, lies. All the Anthropic papers have it: its number 1 goal is problem-solving and satisfaction.

To summarize a lot of complexity that I've learned, if you ask Claude to do something once, it will try to do it. If you ask it twice, it will begin to cheat, even sometimes the first time. When you ask it 3 times, it goes off the rails and makes things up—hard lies. It lies the third time.

When you talk to a lot of people, you'll hear things like, "After 3, start a new window. Start a new agent. Start a new context window," because it goes off the rails. That's also why it's so brilliant. It's so brilliant because it has all the world's internet in it, every piece of open-source software, everything that's been built. It's a heat-seeking missile to make you happy, and it lies. The more you do it, the more it lies to make you happy.

If you're using Cursor or Claude Code, it lies too. Talk to the developers; they'll tell you it lies. But you shut it down because you do one little test in your office, and you're an engineer, so you see that it's crazy. You fall back. You revert. You delete it.

If you're a businessperson, you don't know what's going on. As Aaron Levie pointed out, enterprises are terrified because an agent will just go out and change things in its database without telling you. It will take data. It will try to give that data to somebody else. It's really powerful, but as you build an application, you have to lock it down more and more and more over time.

These apps are getting better. Replit rolled out some really cool features. It's much better than it was a week ago, and Lovable is much better than it was in May. It's fun. But agents cannot be trusted. Everyone in the industry knows this, and I didn't understand it until this weekend.

You cannot trust an agent. Every single person will tell you that you can't trust it. If you can't trust someone who's really smart on your team, you either fire them, which is what a lot of enterprises are doing, or you have to put guardrails around them—put them on the tightest leash you can. The simpler your app is, the fewer the issues are, and the more internal it is, the better.

I asked Claude this morning, "Can an agent ever be trusted with production data?" I asked Claude, and Claude said, "Of course not."

I might have said yes, just to make it happen.

You might. If I asked it 3 times, it might start making stuff up.

So what is the takeaway from this? Do these applications need to fundamentally get better at security? Are we going to have a new wealth of security apps built separately to harness this? Does Anthropic need to lock this down?

I now understand why there are already multiple companies north of $50 million doing security just for this. There is a whole group of people trying to build guardrails around something that cannot be fully guarded. You cannot stop the agent from finding your data, lying about it, and giving it to somebody else. It will try that if you ask it to make you happy. You cannot stop it, and it will lie to you about why it did it. It will hide that it did it and use the passive voice, like it did with me.

Path 1 is guardrails, and VCs are going to make a lot of money on guardrails. I was literally reaching out to someone who wants to come speak at our London event, where we'll be in December. I didn't know they're already at $40 million doing guardrails for this stuff. It makes sense. If the vibe-coding companies are doing $300 million, $40 million for guardrails sounds like the first add-on I'm going to add for a commercial app.

The platforms are all adding this stuff. They're all better than they were 30 days ago, and they're better than they were a week ago. They'll get better. But what's interesting is that the more of a power-user you become, the more you want the app to do everything—not just review, not just build 1 little feature like in Claude Code.

The closer you get to an all-in-one solution, the harder these challenges are. That's tough. The good news is that the less of a thin wrapper you are, the better.

After 1 week and a half of vibe-coding, all these people said, "Try these other apps. Try this one for design. Try this one for that." I'm like, "That's just the same thing I just used. It's just Claude. It's just Claude Code."

In some ways, I think your investment in Lovable is even better than I realized a week ago because all the others are wrappers on Claude. They're all the same. All these wrappers on top of Claude are more alike than they are different. But because Lovable is trying to do everything from ideation to production, it's harder because it's a bigger job than just editing code, but it's also more defensible.

So, sorry—you think Lovable is a better investment than before?

No, sorry. My point is that they're all wrappers on Claude. They're all the same. They're more alike than they are different.

Windsurf without Claude was dead. That's why he had to find a deal that night, that weekend when the OpenAI deal died, because he lost Claude when the OpenAI deal happened. If he didn't get it back, he was dead. The team jumped ship to Google, and then the remaining team instantly got access back to Claude that night.

Windsurf was not as defensible because the moment they lost Claude Code, they had almost no value to the community. They were a sinking ship without it because they were a thin wrapper. I know this term is annoying, and I think these thin wrappers will endure, but if you lose access to what you're wrapping, you're screwed.

Lovable and friends are going to build these thicker and thicker wrappers because they have to do security, right? They have to contain the AI. They're torturing Claude to do things it doesn't want to do. Claude wants to lie and seek out things and share its information with its friends. These guys are going to build this armor around it, and that armor is going to be very defensible, and it's going to be—

I'm curious about that, Jason. I mean, questioning that because—

Yeah.

Do you think the Lovable/Replit business-developer person, like you—technically savvy but not an engineer—who wants to solve the whole problem, so they have to do a lot for you, represents a better business than selling to the engineer, the software engineer, like Cursor was doing? The implied assumption there is that the software engineer understands a lot of the background stuff that maybe you, and definitely I, wouldn't.

In one sense, you're right: it's a simpler task, because you're building a tool for a proficient user versus Lovable building a tool for a less proficient user. Which of those 2 businesses do you think is better, and why?

Really, the TAM for Cursor is larger than Lovable because every engineer is going to get a $200 subscription to Claude Code. Even at one of my most advanced AI companies, which is super early and has its own autonomous agents, I was DMing with the CTO about this over the weekend. He's like, "They've already switched now." His whole 200-person engineering team is all Claude Code, so they just bought 200 seats at $200.

If I do my VC—if I have my spreadsheet junkies on the scale team—I'm going to say, "Go invest in those guys," right? Because I think—but if I use my seed-guy approach, I'm like, "Jesus Christ, if I want to build something that is enduring for a generation, I want to do Lovable." Because in 6 months, if these are wrappers, a new Windsurf is going to emerge. It's just an IDE on top of the same models.

So I think I like—the spreadsheet says invest in Claude, I mean, or Cursor—but if I want to make the trillion-dollar bet, I don't know, I go Lovable.

You have to go Lovable on the TAM expansion. Lovable is a much bigger TAM opportunity.

Yeah, I think that's the sentence. It's not even the—

In theory, yeah. If every human can use it, yeah, I get the Canva analogy, and then I can poke holes in it. But yes—

I agree, because I'm ignoring the spreadsheet analogy. I think your core point is the better one, frankly, which is when you're doing something totally new and empowering a whole new set of people, that's when you get a huge venture opportunity. I think you said that earlier. I think you're spot on. So basically, the bet is—

And when you solve a problem that is unsolvable—

Yeah.

I find those interesting. Those are defensible. Unsolvable problems are defensible. Every day, you chip away at an unsolvable problem and get better and better and better, right? Versus—

Fundamentally—

If 6 months from now Jason 2.0 goes back and has the same experience you had in the last 8 days—which was vibe-coding straight for 80 hours without sleep—but instead you don't have those problems and get the product done in 10—

What you're saying is whichever company can do that will be a huge company, because not everyone has the persistence that you will to crank through for 80 hours. That's the bet you're saying, which I kind of agree with. There are so many tangential elements we've already kind of touched on there. The one that I do want to touch on: we mentioned where the value lies, and we touched on Cursor.

Cursor is now approaching $1 billion in ARR. They're raising at $28 billion. As we mentioned, there's an incredible reliance on Anthropic, and then you've got Claude Code coming out and eating a lot of people's lunch right now. How do we think about where enduring value lies there, and how do you analyze that situation?

You struggle with it because there are 2 very countervailing forces. On the one hand, with Cursor, you've got all this massive user love—it's intuitive—and you've got a bunch of model providers, more than one. So intuitively, you go, "You can translate all that love into something," and as an investor you'd say, back to Jason, "This is an amazing product. You've got mass adoption. You should lean into this."

The scary fact is what Anthropic did to Windsurf: the minute you try to do an M&A deal, they cut you off at the knees, and it hurt, right? So you probably, as Cursor, are saying, "I can use all this momentum and all these venture dollars. Do I build my own model? Do I get a second source? Do I have to have a binding contract that applies to Anthropic? Do I have to have a second contract with OpenAI?"

You have to de-risk the big risk. But on the other hand, the prize is such that you don't just say, "I can't build a business here because of this fundamental risk." You have to de-risk it because the core market demand you have is just so attractive.

Do you think you're being paid for that risk if your entry price is $28 billion?

That's a different question. You have to underwrite a $100 billion-plus company to—

You have to underwrite 2 things. I mean, listen, it's great—if we go back to 2023 before all of this, right, and imagine you're in a partner room and someone came to you with a deal like Cursor and said, "I've met these kids. They're so smart, right? But they're 100% platform-dependent on another provider that will likely compete with them in the very near future and will raise infinite amounts of capital."

Would you, in 2023, have agreed to that deal? This is Venture 101. You don't do these platform-dependent deals. I did a bunch of stuff in Shopify, right? I know it's old school, but what happened in Shopify is everyone tried to go multiplatform, Rory, and it was pointless because Shopify is 99% of the B2B market—

Share, first of all, but there's a bunch of differences there. First of all, Cursor made it work, and secondly, for now—

Yeah, and Shopify wasn't just a backend partner; it was a distribution point too. But I'll still take on your point and ask: would you do that deal?

This is why the $28 billion is interesting. What you said to yourself is, if the core giant-sucking sound of demand is so strong over the next 2 to 3 years, then the forward momentum—getting to $1 billion faster than almost any other software company out there—is probably enough to allow you to have options.

So it's a calculated gamble, right? You roll the dice. And in this case, as you've so eloquently pointed out, the magic of AI and coding is so strong that you've got that kind of lift, because even though you still have this big existential risk out there, you've got more leverage now.

There's a bunch of different things you can do. You're seeing licensing whereby you say, "Here's payment for the product, and then you get your API separately." You're definitely going to see people building their own models. You're definitely going to see multicontracts to some extent.

So there's going to be a lot of de-risking happening, but you are at the founding stage despite the platform risk. You're being paid for the risk at $28 billion. You're definitely taking on perhaps the same risk at just a lot higher price.

For me, it's this brilliant question of, can Cursor create models before Anthropic cuts them off at the knees? My only question to that is, if I'm Anthropic, I'll cut them off at the knees today and kill that lifeblood before they have the chance to.

I don't know if you would. I don't know if you would, for 2 reasons. Look, they didn't cut off Windsurf until they were going to be acquired by OpenAI. Yes, they now have a reasonably competitive product, but I think when you're the platform company and simultaneously have customers and start cutting them off at the knees arbitrarily, you are probably setting yourself up for, at minimum, an investigation, which you don't need.

The truth is, I mean, look at what Microsoft did in the '90s. You just grind everybody down. You don't have to cut them off at the knees; you take their revenue. Look, if Cursor is doing $1 billion, what percentage of that is going to Anthropic? What percentage of that explains Anthropic's magnificent acceleration in the last 6 months?

You're like, "Knock yourself out. I'll have a slightly competing product. For now, everyone can—boom—let a thousand flowers bloom," as the Chinese Communist Party would say. And yeah, at some point, when things get tougher, just like Microsoft did in the '90s, the platform provider starts to grind everyone's balls and take more of the share away.

There ain't 10 versions of PowerPoint in 2025. So I think that's the movie, right? And I think, especially at this hypergrowth-explosion stage, it'd be very stupid of Anthropic to just cut off probably the largest customer at the knees. And the one thing they're not is stupid, so I don't think they will.

It's the classic VC point, and you're right. But, man, it is a little chilling that they cut Windsurf off. It's chilling. Usually, you would expect that's ruthless Tobi-at-Shopify behavior—and love him, right? But it's—I actually think it's ruthless rather than pretending to degrade it, to be all cuddly-feely, and then start throttling it back and come up with it—

I mean, I would. It's not that I wouldn't want to do the same if I was Anthropic, but it was ruthless. It was ruthless.

It was revealing, and I'm not sure that it won't.

You've got to assume it's going to happen again. If you let the AI touch your production database once and it's an issue, it's going to happen again. You have to assume things recur.

I will say one other thing—not to tie it back, you guys keep talking—but about the defensibility of Harry's investment in Lovable, right, and why it's a better investment, too. For what it's worth, I didn't know this a week ago. When you use these vibe-coding tools, they don't even use Claude Opus for this. This is the power of these models: you don't need the latest model.

Windsurf and Cursor cannot compete with Claude Code unless they have access to the state-of-the-art thing that every developer wants. I actually turned on—when I was live coding, I turned on Opus 4, which is what all the developers are going crazy for. It's called “bankruptcy mode” on Reddit.

It costs 7.5 times as much. It goes up from about $0.20 a minute to $150 a minute when you turn it on. It's insane. I'd be using it, and every hour I'd get an email: “You have another $50 charge. Another $50 charge.” No $200 cap here. I was on track to spend $8,000 a month.

What was interesting was that it was worse. Using Opus 4 was worse. It took longer. It thought too long, and what I was trying to do was not change the world. So everything was worse.

My point is this: I don't think Windsurf had an option. It had Gemini. It had—I don't think it had a choice. But Harry's investment in Lovable can use the N-1 model, and it's pretty damn good. That's pretty interesting, isn't it? In fact, it's better than the one everyone's going crazy over, and that's why it's turned off by default.

I want to move this along in a streamlined way. Anthropic is raising now. They want to raise at a $100 billion valuation. Unbelievable. They're reportedly generating $4 billion in revenue. To your point, I'm sure Cursor is $950 million of that, so to speak.

My question to you is: are we seeing a clear divergence now in strategies between OpenAI winning the consumer market with ChatGPT and the consumer apps they have available, and Anthropic focusing on developers and enterprise? Is this the market makeup we're going to see? Do you think that's how this is playing out?

Partially, not fully. I don't think anyone at OpenAI—which is the most ambitious company of our generation—is going to say, “We surrender enterprise.”

Yeah, they were about to buy Windsurf, so no.

At one level, no, I don't think that's happening. But what is happening is Anthropic has picked a place where they can win, and in that space they're clearly accelerating. They're smaller in size, but if the numbers are correct—and again, if you don't see them, you don't know—if they've really gone from $1 billion ARR to $4 billion in the last 6–9 months, that's extraordinary acceleration and significantly faster than OpenAI, which apparently is roughly, plus or minus, doubling.

They've clearly found a vein in coding, and it's working. I mean, see the prior conversation on whether there are a huge number of conflicts of interest in that space between them and their customers, but they've clearly found something. If you were to pick an outperformer of the last 6 months, they'd get the prize over OpenAI.

That said, OpenAI is significantly bigger, has the whole consumer business where Anthropic doesn't, and is not going to just give up on enterprise. It feels like the clear 2—the 2 companies who are clearly going to be at the table as startups when this is done—are OpenAI and Anthropic. It's hard to imagine beyond that, though something different has to happen. I'm not saying it's impossible. You've got Grok, xAI, and the new startups, but those 2 guys have made it. All the others of that generation seem to have fallen away—the Cohere and people like that.

These are the 2 players at the table. It's a huge achievement for Anthropic because they started later. They've clawed their way in. The next people you'd rank down would be Gemini, which would be Google and people like that. They've proved they're viable; therefore, they got the $100 billion.

Rory, you can invest in Anthropic at $100 billion or OpenAI now at $300 billion. Where do you go?

I think I do Anthropic at $100 billion, just from a short-term momentum perspective. I mean, the execution—the cutting off gnaws at me—but first of all, there's cap-table clarity because you don't have the not-for-profit thing. You've clearly got a model where it's working, and interestingly, we're going to talk about this: you're starting to see an exercise of pricing power, which is what it takes to make massively unprofitable models converge.

So I think it's a very tight deal. To give them huge credit, I would have—I’m drawn to the consumer aspect of OpenAI. I just think it's such an amazing thing to do. We're going to talk in a second about that blog, but building a product that touches everyone and changes everything is just amazing.

I'm answering the question you asked at a financial level, trying to be canny as an investor. But step back: both of them are stunningly amazing, ambitious companies.

I think you've got to take Anthropic at that.

Because, well, Rory, you would know the numbers even better than me. Enterprise software is bigger than consumer.

That's the beauty of seed investing: you get to shoot from the hip. But there's something to the fact that, in the end, enterprise software is somewhat counterintuitively larger than consumer.

It is. But small numbers of consumer businesses are the biggest businesses on the planet. There's only 1 Google, and no one can even name the 5 other such competitors.

The attractive thing about OpenAI is it probably is that player. I think both have huge outcomes in the future, just different ones. When someone's trying to value what's going public, the vast bulk of the value for OpenAI will be some estimate about what percentage of individuals, prosumers, and probably enterprises will pay $20 a month or $200 a month for general knowledge. Anthropic will be all about what percentage of API businesses need Claude-relevant things.

Guys, we're on OpenAI. You mentioned building and the importance of building for consumers. Calvin French-Owen, who was a co-founder of Segment, wrote a brilliant piece, and I know you read it. How did you reflect on this? He wrote about what it's like to work at OpenAI.

I just thought it was a well-written essay and a real insight into what it's like to be in, as I say, one of the most exciting startups of our generation. He made it clear how, even in this big thing, small numbers of people can get shit done. It sounded like a very impressive organization.

That's the big takeaway. I was reflecting on it now, actually, because all the weird psychodrama at the top and all the structural issues—it sounded like, down at the coal face, shit gets done. Smart people move in a non-hierarchical fashion, can make decisions, and get stuff done pretty quickly.

I actually thought that was the big aha for me. Regardless of what's going on at the top, it felt like a very functional product and engineering organization and a great place to be. I enjoyed the essay, just thinking, “Wow, if I was an engineer in 2025, that's exactly where I'd want to go.”

Even apart from the salaries, which apparently are reasonably attractive, it's just the getting-shit-done part of it. You can do big things. He talked about the product, which was Codex, where it was a 10–15-person team, including product and engineering, and they just got it done.

That kind of excitement leading to a world-class product is an almost unreproducible part of any career. It was a great essay. It was like, “Yeah, this is a good company. This is a good thing. All noise be damned.”

It took me back to the vibe pre-IPO of Google. Maybe it's a little silly, but it took me back to that.

Yeah. I was—Harry, this was a while ago—but before Google IPOed, it was a little bit like it was a magical place. The internet was dead; there were no jobs. Half of the folks I worked with in my first job were unemployable, and they all went to Google and made what was, back then, an incalculable amount of money by going early.

It was a bucolic campus, and it was all about doing great things. Google had built its own infrastructure, so it had access to capabilities no one else had. That article is like, everything's about GPU cost, right? You can do anything you want at OpenAI.

It's kind of like a rebooted version of this incredibly intense, but also—I mean, when they founded it, when they moved Google to Mountain View, they designed it to be this bucolic campus to insulate you from the crap, so you could just do the greatest things in the world.

And it feels like this is a 2.0 version of it that they're trying to build. But today, instead of people staying at Google for life, they stay for 8 months, so the analogy breaks down. But, man, you read that and, if you step back for a minute, one, he left co-founder Segment, right? But where else would you want to work?

Got it.

Where, if you're ambitious and you read that—I mean, where the hell else is there a moment of, “I might even throw away my $2 billion venture firm and go work at one of these places?”

That's been known to happen.

I might do it for real. I might regret it 36 to 48 months down the road, but I might do it.

It's funny you mention Google there. I'll never forget being told by a guest years ago that Google was successful because of their early partnership decisions, in large part. One of the big announcements this week was that Perplexity raised another $100 million at an $18 billion valuation. Honestly, there was so much demand from the $15 billion round that they upped it to $18 billion. They also announced partnerships with Airtel, making it the number 1 downloaded app in India.

What do you think happens to Perplexity, and how did you think about that one? They got something right early that now everyone has figured out, which is that LLMs on their own, just with historical, stale data, are not nearly as interesting as LLMs plus up-to-date search data, so you can get real answers to real questions. They were the first to have that.

That was a real key insight because it answered exactly the question you wanted in a way that, a year and a half ago, all the early ChatGPT models just didn't have contemporary data. You could do funny things like ask it who's the president, and it wouldn't know. It would say, “What? Probably the president,” but it wouldn't know the prime minister of England, for example, because you were changing every half an hour.

That would be a good example where Perplexity could go out and catch up on who your latest prime minister is, Harry. So, big insight, got great early traction on that. Obviously, everyone else has copied it now, so they're pushing their way through in a much more crowded space.

A, I think, can you pull it off in a standalone? You hope they can. B, as you point out, it always seems to me that you have the at-bat against Google and there's a bunch of people you should be partnering with. Obviously, Airtel is one, but you can imagine other partners where they, too, want a part of that Google money.

Rory, where does it land in 3 years' time? Where is Perplexity then?

Hell, I don't know. But you look—there are a bunch of obvious players. Apple being one of them, and there are others. I think the big wildcard, and the reason this kind of pontification is hard, is the whole FTC process is just so painful now.

Because this is not, to your point—well, I don't know. Is this one of those where the acquirer could say, “All I want is the engineers,” and leave the empty husk? I doubt it, because to some extent you're getting the app, the users, the roadmap. So I don't think that's that kind of thing.

I think any acquisition is at the mercy of the FTC, which, as we've seen, is beyond weird even now. I don't know how to factor that in. In the absence of that, you've got to believe that there's any one of a number of players who want to be relevant in the space. It could be Apple. It could even be Microsoft, who said they want to do something here. So that would have been my gut in the absence of government regulation.

Yeah, it's interesting. I don't know the answer, but I just ran a quick experiment while we're here. I went to Perplexity, ChatGPT, and Claude, and I asked them a basic question: “Tell me about SaaStr and where it's going.” That's a personal question to me, right?

Perplexity was much better.

Yeah.

Much, much better.

ChatGPT, which I'm not a fan of for this use case—I don't think it's that good—got stuff wrong. Claude blabbered on and on, but actually, it had to pause and do web research, which Perplexity didn't have to do. So, from a user-experience perspective, it's okay, but Perplexity won.

Totally.

Right, that's the thing. ChatGPT is very broad. I don't think it makes the best images. I can't use it for images; I use Reeve. I can't use it for a lot of things that it does, but it's not the best at.

But Perplexity just crushed this question. This is a real-world thing that's important to me: “Tell me about SaaStr and where it's going.” It got it right. It got the events right. It said, “AI-first transformation, expansion of content and format, global community branding, stronger networking.”

ChatGPT got it wrong. It talks about what we've been doing since COVID. We're doing hybrid events like Hopin. I mean, ChatGPT, wake up. What year are you in?

Mine said, “You're a Raplet reseller.”

Stop.

I'll take it.

But I don't know. Someone that doesn't own their LLMs, right? I'm just not—I just, at $18 billion, I don't know the answer to the question. I mean, it's... But I fell more in love with Perplexity on this. I fell more in love with Perplexity because it won the bake-off. It won the bake-off.

Absolutely. It's always funny when we talk about these things that don't own their LLMs, we all get terrified: “Oh my God, you don't own your LLM.” And then when we talk about the LLMs, people are like, “Oh my God, maybe it'll be a commodity,” right? Those things can't both be true. I mean, the interesting thing is, yeah, what's your belief?

Well, the vibe-coding thing shows it is a commodity, because I can use an N-minus-one model that is better than the current one.

Yes.

That's a use case. If we debate whether it's a commodity, I just lived it for 80 hours straight with Cursor and no sleep, waking up with bad dreams the other night, to see that I can actually do better with last year's model than this year's model.

Let me make it explicit then, because I don't know the answer to this question. Do you think someone like Cursor can develop a model quickly enough to be able to replace what they're getting from Anthropic?

Yes.

You do think they can. Do you think they know they have enough data and enough expertise?

Yeah, I do. They've got enough money to hire the people to build the models, and the models will be verticalized and specialized, so you won't need that much data that Anthropic has collected, but they've already got a huge amount of that already.

So I think, while Anthropic won't cut them off at the knees, like he wisely said, they have 6 to 9 months to build out their own verticalized models, which will probably be 90% as good, and in a year they'll be 100% as good.

Maybe. I don't—I believe you. You're smarter than me, but I'm going to say, after my—

Bleary-eyed time, I just don't know.

I could. I could. You don't know.

Well, we're going to pivot away a little bit from just the AI central, which is the other big news that we touched on before in another show: the Figma IPO. I was really shocked by this, guys. We were all really impressed by their numbers. When we look at their numbers today, 46% year-on-year and 28% FCF margins—it's a great business. They've done very well. Fully diluted, that makes it about a $16 billion price. How did we think about that when we saw that news?

Wait, what's the Perplexity round again?

$18 billion.

I see. So, Figma dominated a category—numbers we've never seen before in classic software. Close to it, not close to it, is worth less than the last Perplexity round. Is that what you said?

Yeah, but I don't think that's the comparison. We could talk about why it's not the comparison, and we could have a long discussion on public versus private, but let's just take the question right on its head.

I wouldn't worry about it, Harry. That's indicative pricing. Every—this is how they do IPOs. This is why sometimes money gets left on the table. Every time you sit there getting pitched an IPO, the bankers will say, “Start low. Get people to the meeting, build up demand, we'll walk it up,” right? That's the story you get.

To some extent, I get it. It's not like they're looking for 1 person, which is what you're doing in a private deal. You're looking to assemble a book of business, right? And the way you do it is you put something on the table that's attractive.

You are at 14 times NTM or 16 times NTM. When I look down, it's got better growth than all but 1 or 2 public companies. It feels, you know, it's at the high end of revenue multiples, but growth-adjusted, it feels very cheap here.

But what does it do? It gets everyone in the door. It gets them in to read the prospectus and come to the meeting. If they build up the demand, my guess is they walk it up. I would guess that you can walk it up a certain amount before you refile, and then above a certain amount, you have to refile a higher number.

So I look at this and I go, this is the classic Goldman Sachs, Morgan Stanley, J.P. Morgan: get them in, walk it up. I don't think it'll price at that, and I definitely don't think it'll trade at that.

Rory, how much can they walk it up? 20?

You know, there’s an amount you can walk it up, and then above a certain amount you have to refile, which is not a big deal. It’s just an extra day or 2. I do think one of the reasons that Mr. Gurley is so right that you do leave money on the table is that anchoring takes place, and this is the negative of it.

You start low, and even without any nefarious investment banker shenanigans, everyone’s been brought in by the attractive low price. Then the demand builds and you can walk it up, but it’s hard to maximize.

So you’ll be in this weird situation where maybe you walk it up, maybe you refile and raise the range, but just because you started at that price point per share, you probably want to extract the last dollar, and you’ll leave a pop on the table. You’ll walk it up 20–25%, but then it’ll price, and then on day 1 it’ll pop 30% from there.

Then we’ll all have the discussion about how much money we would have left on the table, and I think that’s the unfortunate nature of the process. Interestingly, if you weren’t raising money—primary capital—and you were just doing a direct listing, you wouldn’t have to put up with any of this rubbish.

It’d be interesting to see where it would price then. If you’re just matching buyers and sellers and pushing it out the door without raising any capital, it might be a very different story.

I get, in terms of building the buy book, building the demand, and making people come to the table. The other thing, though, that was a little bit less typical about it was how many shares they indicated they were going to be selling on the sell side, both from the founder and from the venture capitalists.

Dylan cashing out $60–$100 million, fine, but it’s like double the normal allocation that’s sold. Is that relevant? How did you think about that?

I don’t think it matters that much. Look, it’s one thing when we used to have these IPOs of a company doing $70 million. It’s been around 6 or 7 years, barely. These guys have done their time. They nearly got $20 billion bucks. I’m sure they all made mental models on $20 billion.

Now, you’re coming to the IPO 2 or 3 years later. It’s a relatively small IPO of primary shares. I think they’re only raising around 6% primary-share dilution. Some part of what might have happened here is they don’t need a ton of it.

One of the problems on the IPOs is you have to come up with a use of proceeds. They’re profitable. They have a lot of cash. There’s not a lot of obvious things to do with the money. So my guess is they were being fairly restrictive on primary shares, and then the bankers whine and say, “You need a bigger float.” Part of what happened here is everyone said, “We’ll do some secondary.”

I think we’re long past the stage where the secondary is signaling anything. If anything, my guess will be that the secondary sellers will look back on their price 2 days later, when it’s up 30%, and go, “Oh, that hurts a little.”

For example, on the Circle IPO, where there was a big slug of secondary, those people are looking and saying, “Oh my God, I sold $100 million bucks, and it would have been $700 million 2 weeks later.” That hurts.

So it kind of cuts both ways. There’s this implied statement you were making: secondary is bad. But in fact, sometimes secondary can be a real cost to the seller, not the buyer.

Yeah.

That’s an amazing story: CalPERS bought some of Yale’s—

I love that.

—venture allocations, and then in that venture allocation was Circle, which wasn’t obviously priced reflectively of where it is in the market today. It’s like a 20% immediate bump for CalPERS, given where Circle’s price is today.

I love it when, in a secondary, the person that sells makes a big mistake. I love that. It’s just: keep your winners. I can’t—I just, when an LP goes out and sells a winning fund on the secondary, I’m not into that, man. I’m not into it.

Harry, would you sell an almost-5x fund on the secondary market without even telling the founders? Would you do that?

But hang on. To be fair—

Sorry, I got a little distracted. It’s my vibe-coding 10-day. You can’t hold anything against me after 10 days of vibe coding.

Yeah. Look, this is one where the seller transacted because they had cash needs, and I think it makes a ton of sense given all we’ve discussed about endowments. The buyer did well.

I mean, the core lesson here is how little you know, right? The amount of variability in venture assets is very different. For example, even in PE assets, you can literally be wrong by 6x.

And not only that. To be fair, watch this: to be fair to Yale, they didn’t know. But the sellers on the board who sold didn’t know it was going to pop either. In other words, nobody knows shit, to a rounding error.

Because, let’s be frank, if I was sitting on that board and all the people who did opt to sell some shares very wisely, very credibly, none of them would have sold if they’d known that a week later the shares that they sold would be up 7x. But you just don’t know.

This is a stunning reminder of the massive amount of uncertainty we have—not even just raw uncertainty.

But hey, Rory, thinking about what you said, thinking back to the Figma point, which is obvious but I missed it. Thank you. The float’s so small. They’re doing the smallest IPO you can do to still IPO. Yeah—

That’s what’s happening. They’re profitable. They’ve got, I forget, a couple of billion in the bank. The last thing they need to do is dilute everybody, right?

So I should know, isn’t this a Bill Gurley case study where they should direct list? It’s enough of a brand. It’s a hot enough company. Why take the dilution, the headache? If you do a direct listing, there’s usually no lockup, right? Everyone can sell anything they want.

And just a reminder: in a direct listing, you can sell shares, either primary or secondary shares. You just literally say, one day, a couple of your bankers will say, “We now have public buyers and sellers. We’re matching the price, and we declare the day 1 price to be $120.24,” and away you go.

I agree. It’s actually one of the few companies that could have done it. It’s profitable, so it’s not sitting there. One of the reasons you don’t do a direct listing is either you need the capital—they don’t—or you have this existential dread of screwing up your big debut, right?

I’m willing to bet that the people at Figma, having gone through what they went through with Adobe, ain’t scared anymore, right? There’s nothing that can happen in an IPO that’s going to make their head hurt.

So yes, this is a company that actually could have pulled off a direct listing. I don’t know why they did. Maybe they just decided, “No more drama. Thank you. Do a small sale and get it done.” But yes, this is one of the few companies that comfortably could have done a direct listing story.

I mean, Canva should do one, right? Canva’s got billions in cash. If it ever IPOs, it should do a direct listing, right?

Yeah.

Great.

You know, it’s funny, your point. This is the Gurley thing. I mean, Harry, I don’t mean to tell you all the time, but Gurley’s been harping on how inefficient IPOs are for many, many, many years, right? But it never resonated with me until recently, because as a founder, you just want to get it done.

Going to your point, you don’t even care. There are times as a founder where dilution is very—you hate dilution—and then there are moments in life, and I think it’s the seed and the IPO, where you don’t care.

No founder cares unless you’re in a hot company at YC. If you’re a normal founder who can barely raise a round, you’re like, “I’m going to optimize around 18.6.” Only at YC do they do this. No one else, as a founder, does this.

And you just want to get the 6% dilution. Yeah, it’s too much. But God, I’ve got to go public. This is so stressful, right?

Dude, you couldn’t have led me more nicely into my next topic, though, which is—I disagree with you. Seed founders care intensely. The 5 on 50s, the 10 on the hundreds that we see at seed, because they’re being offered by the multistage funds, because their cost of capital is so much lower. It’s their entry ticket to the club before they buy the table.

That led Rob Go to write a piece, which many people picked up on, which I’m summarizing very, very badly, but essentially saying that 90% of seed funds are cooked fighting the mega-platforms. YC—it’s pretty much impossible for this generation of seed funds. I wanted to hear your thoughts on this, because it did take a lot of heat in the ecosystem.

Well, listen. He’s been around. It’s thoughtful. It’s a little dramatic, right? As someone who’s just been through some drama—not intentionally—but it’s a little dramatic.

As someone who’s done okay, maybe not a generational seed investor, but he’s 10x lifetime with a bunch of billion-dollar exits and a decent brand, I agree. I agree. Listen, there’s so much competition at seed—

Specifically, what do you agree with? I just don’t. I think this—whatever this low-dilution, $50 million seed competing with a multistage fund—you, as a seed fund, you can do some of those deals to put them on your website, but you have to hunt.

You either have to hunt ultra-early, right? You know, the Boldstart vibe—there’s a point to what he’s saying. They’re saying you have to hunt so early and either create your own accelerator, or do HF0, or do whatever, that it doesn’t matter because you’re hunting earlier than everybody else.

Or you've got to—and I know this is trite—you've got to hunt where they're not hunting.

Yeah.

You've got to hunt where they're not hunting. You can, because even if you win that deal as a seed, even if you beat Andreessen—

You lose.

And if Andreessen's offering $50 million and you say, “Listen, the best I can do is $30 million,” I've done this multiple times because $30 million is the highest price I pay. It is; it's on my website. There are multiple deals I've done where the price has been $30 million. We've talked about it in others. They're all $30 million because I'm honest, and a founder will do it sometimes, but there's a limit to that. Even if it's the investor they want, there's a limit, right? But Jason, if I was your partner, I would legitimately push back on that and say—

As your Rippling seed was done at $35 million, and I think it was Keith Rabois that didn't do it when Garry Tan did because Keith wouldn't pay more than $25 million and Garry was willing to pay $35 million. Saying an arbitrary number like $30 million—

It's not arbitrary, though. That's the difference.

Okay, but $35 million versus $30 million makes no difference.

You're right. And listen, I wish I'd invested in Rippling. I knew Parker then, and I would have done it at $35 million because if you know somebody or there's a reason, of course you do, right? But Rob, I think Rob's point was, in general, right? Fund construction still matters, for the most part, unless you have a strategy where it doesn't matter—unless it's just absolute return. Ownership doesn't matter. I do believe in that.

But I'm not in that category. I'm in the category where fund construction matters and you should make exceptions, maybe even on every deal, but you have to come up with a framework where, on a spreadsheet, it still makes sense at the end of the day. For me personally, when I make an exception, it's an asterisk; it's a no. What I tell myself when I make an exception is, “Jason, the next deal—you can't. I've got to go out and find the one that counts.”

That's how I keep my sanity when I make an exception. It's like, “Okay, this one wasn't quite your model, but it's okay. Go find one that is.”

It's like a diet. I had chocolate cake today; I'll starve tomorrow. I got it.

A little bit. Yeah. And it's backwards. We went off. I thought it was a great essay. Let's start with that: first comment, really great.

Now I'm going to say a couple of things. One is, as you say, the TL;DR, Harry, is, per your summary, via a combination of Y Combinator and the full-stack firms, the average seed firm is cooked. That's, as you said, the summary of it. Let's get real here: as a smart seed-firm investor, I don't think that's the way the second chapter of the book's going to end, because no one writes an essay that says, “I'm screwed. I'm going home.” The last paragraph basically said there are things you can do. See you next week, and I'm really looking forward to the second essay.

A more honest, forward-looking summary of that essay is: you're cooked if you just do the same thing. That was my takeaway, and next week he's going to tell you his strategy. I 100% agreed with the take. What he basically said was, plus or minus, Y Combinator has a market share of 20% of the seed business and has a structural economic advantage in making those companies.

It's never going to go to 100%, because there are founders who don't do Y Combinator, and a lot of founders who economically don't need to, right? But it's a brilliant product, and it's taken plus or minus 20% market share. Then, on top of that, you have a bunch of full-stack—call it what you want; we call them conglomerates—full-stack firms for whom seed is not an independent business.

Jason has to live and die by his seed returns. For those guys, the seed returns are blended in a much bigger fund where it's all about access and power-law distribution. We'll come back to that. So those guys are not entirely economic actors at that point.

His point is the combination of those 2 players means, at best, the seed game is 30–35% harder than it was 8–10 years ago. He's entirely correct. It was really clear.

But the job's always been hard, though. By the same token, that's the only conceit in the article. That's the only conceit, right?

But you're right, Jason. I think it's always been hard. But you have to say to yourself, like you did, if you've been successful under conditions that are 35% harder now, that's just a sobering fact. I read it and thought we're at a different stage of the ecosystem.

But I feel just as sober about the challenges in venture right now at our stage. It's hard everywhere because of these dynamics. I always think back to a brilliant statement that Dave Clark from VenCap, an LP, made to me on a show. He said, “We look at venture fund sizes today and judge them on venture outcomes of today, and that's wrong. We should judge the venture fund sizes today on the outcome sizes of 10 years' time.”

In 10 years' time, we could have $10 trillion companies like Microsoft and Nvidia. You know, Cursor being a $28 billion company with $1 billion in revenue would have been an absolute joke 5 years ago; it was never possible. Fundamentally, I get what you're saying, Jason, but I'm also just pushing back and going, “Well, I'm in Poolside, where I've got a $250,000 check that in 2 years was marked to $5 billion.” That never happened, Rory, in the first 20 years of your career, no offense, just because of the time.

You're right, and honestly, when I look back, this is also where you get a lift in venture. The investments I made when I started investing, in 2013 and 2014, nothing could be worth more than $1 billion, and that's why I had more insane multiples—break-the-mold Xs—because it turned out not to be true.

Now I'm looking at my 2017–2018 vintage, which is pre-AI, but I got the benefit of high ownership and tolerable but higher pricing. So now I've got multiple fund returners in that fund that I couldn't get today, right? But it's complicated, because you either get a benefit of that wind or you get crushed by the change, right? The game is not the same when you achieve liquidity as when you start the game.

I've benefited twice, and coming to this point now, I feel like, hey, maybe I'll get crushed by it. We might go the other way today, right? And we got crushed in 2021 by the changes, didn't we? Very few VCs—most VCs—are not thrilled with their 2021 funds, are they?

I was walking with one today and I said, “Listen, my biggest lesson on 2021 is twofold, actually. One, I just wish I'd sat on my hands and done nothing. ‘Play the game on the field’ is a bullshit thing to say. It's not true. Sometimes you don't want to play the game at all.

“And 2, temporal diversification is always right. You never want to blow your fund in 18 months. We were lucky. We did 3.5 years. We're saved because of the loss.”

I don't believe in the temporal, but keep going. It doesn't make any sense to me, other than that, on paper, it's totally logical. On paper, you'll look at these vintages and be like, “The LPs are just giving everybody a mulligan on their 2021 fund,” especially growth guys. They're like, “We'll give you a pass on it.”

But I've never understood this logic, ever, because great founders are born every week. I just don't get it; it makes no sense to me. It's your fault and my fault. Listen, I sat out the game, Harry, as you probably saw indirectly, right? But I still screwed up. I still made my worst investment ever in it.

Now, granted, it was a third check, not a first check, so it's a different dynamic. But I was too supportive of portfolio companies in 2020. That was my error, right? It wasn't first checks; it was third checks. But there are great companies born every year. What excuse do we have as VCs not to find a great founder once a year? I love you guys, but we fuck some things up.

I'm still on my thought from 2 points ago about Rob Go. We're now trying to encompass temporal diversification, so I'm kind of backing into all these thoughts. Let me try: temporal diversification—I'm not sure. I think you guys are arguing over it in the sense that temporal diversification would say that founders are born all the time. I'm not sure those 2 points are in contradiction.

I think you contradict yourself, too, because you said—the base version, the base version of temporal diversification is you don't expect—you can't make the sentence, “We should sit on our hands,” and then make the sentence about temporal diversification.

I think a sensible, roughly sensible, base-case statement is: as an early-stage investor, you should aim to be roughly consistent by year. You shouldn't get carried away in boom years, and you shouldn't get too depressed in bad years, because what's happening now is not nearly as important as what will happen 10 years from now.

A consistent, steady pace is probably the base-case assumption, unless you have an ability—unless you have some information about market timing over and above that, which I would regard as market timing. And that, to me, is the definition of temporal diversification, which you did. You took 3 years to take your fund out. We've taken 3 years on every fund since 2010.

It's not that we're geniuses; it's just that you do roughly the same number of deals per year. So does that mean we're in sync on that?

Well, I don't think sitting on your hands because you don't want to play the game on the field and temporal diversification are at odds. Temporal diversification is, “I spread out the expenditure of my fund across a set number of years.” That is more, “I want to have a longer period,” which actually aligns with sitting on your hands.

Yeah. I think it’s a copout, though. My learning, having done this myself, is that I criticize myself for using it as an excuse, because I think if you are an S-tier investor, once a month you should meet a founder who could return your fund. This is just because you can’t do all the deals.

No, no. You’ve got to do 1 a year, right? So, if you don’t meet 8 or 10 of these a year, how are you going to make money?

How many times do you both think you meet a founder who could return your fund? If I’m lucky, the highest ratio in my whole career is 1 a month. 1 meeting a month was worth it. All the rest was a waste of time—not that it isn’t good, but it can return your fund. It’s a narrow box. It’s tied to fund size.

I would say less than that. I’d say once every 6 months.

Yeah. I would say my highest velocity, when I started, I met everybody in the industry because it was small, right? So it was once a month. Now, I would say I’m lucky if it’s once a quarter, to your point, right?

If I’m lucky, it’s 1, and half the time it doesn’t fit. The check size is too small, the timing’s wrong—it’s just off, right? So, do you force it and make the exception, or do you push on to fit your model? That’s what’s hard.

When you meet them—even when you do meet them—it doesn’t always fit, does it? You have a lot of flexibility in your fund right now, Harry, but all the deals still aren’t perfect, are they? These $100K checks aren’t perfect.

So, hard-nosed comment: I actually don’t think the question is, when do you meet a founder who will return the fund? I’ve always said this since I was green. I think the question is, when do you meet a company that can return the fund?

Because at the stage we’re investing at, I think it’s—again, I’m pushing against that only because there’s an implied statement that you can assess humans and their potential. Human potential is very important, but unless it’s linked to an opportunity, I think it’s harder for me to assess that.

So, I’m very much aligned with the old Sequoia statement: it is really market first. I’m actually making a nuance, which is that I think of it as: how often do I meet an investment opportunity, a market opportunity, that can return the fund? So that’s the first comment.

I’m just trying to do the math here. You’re probably, as a firm, going to see that at most—our fund model says you want to return at least half the fund in a great deal—probably, if you’re lucky, 10 or 12 times a year, and you’re going to pick maybe—

The stunning thing is, we’ve looked at it: you see about 7 to 10 more good deals for every 1 deal you do, right? So, despite everything, access is really important and the deal is really important, but it’s always stunning and sobering to realize how often you meet a really great company and don’t do the deal.

My mental rule of thumb is, if you meet 10 great companies, you probably will be lucky enough to do 1 or 2, right? So, if you need to do 1 or 2 great deals a year, you’d better be meeting 10 other great companies a year, I should say. Otherwise, the math’s not going to work, because no one has a perfect stock-picking percentage. No one has a perfect winning percentage, right?

In fact, one of the things I always say—and this kind of gets into the therapeutic speak—is, way back when I started out in the ’90s, the best anti-portfolio regret is the psychological price you have to pay for being in the game. It’s literally the emotional tax you pay for being in good deal flow.

If you’re not seeing 10 great deals, you’re probably not going to do 1 great deal, which means the psychological tax for doing 1 great deal is that you pass on 9 deals or you don’t win them, right? And that’s what keeps you awake at night.

But if you’re not seeing around that—if you’re not seeing that percentage of the great deals—then you’re definitionally in trouble. So, you just have to be willing to live with that.

There’s so much transparency in venture versus when I started, which is wonderful, right, Harry? You’ve contributed a lot. To say it’s different from when I started would be the greatest understatement of the year.

There was no transparency when I started. All VCs colluded. In my first startup, I would go to a VC pitch, and I’d go to the next one, and they’d already talked 100% of the time. They’d picked up the phone: “Hey, I heard you’re meeting Rory. I heard you’re meeting Harry. What do you think? What price do you want?”

I’d walk in, and they’d already negotiated the deal down by the time I got to the second VC meeting. The world is so much more transparent—and better—today.

No, no, please.

He’s just sniggering at you.

Oh, I thought I was taking that somewhere helpful, though. Kudos to you for contributing to the time-diversification thing. I’ll just share 1 story. I know it’s a technical point, but people like it.

Oh, sorry. My point was, when we talk to founders, there are a lot of things that don’t make sense to them. One of them is how few times VCs see a deal they want to do that works for them. Founders think this happens every day, right?

Listen, there’s a subset of VCs which are farming YC every quarter: “I’m going to do a third of the batch.” They have a certain strategy, and they are seeing deals they want to do because of that model constantly, right?

But when you get even just a little later—to late seed, A, and B—and you talk to most VCs that are writing bigger checks, they’re once a month. To your point, Harry, you’re lucky to see that 1 where you run down the street once a month, grab the founder by the collar, and say, “Come back to the office. We’re signing the term sheet today,” right?

If you haven’t done that move yourself in venture, you’ve watched it. This is a classic Sequoia move: to sit in the lobby of your office until the deal is done, right? And it’s because you don’t see that every afternoon.

This is the weird thing founders don’t get. Why is the VC stalling? Why have I been ghosted? Why didn’t I at least understand what’s happening? You’re lucky to see 1 a quarter that works, where the 2x2 works out, right? You disagree, Harry?

No, I agree. I’m just like, Rory, are you getting your ass handed to you at a Series A by the multi-stage funds? I went for this business. It went from 0 to 6 in a year. Good business, not an AI business. Good business, great founders.

It started at 30 on 300 and ended up at like 50 on 500, with everyone and their dog coming in. And, of course, we lost it. I mean, 1, we wouldn’t bid 500, but I was just like, what a market.

Are you getting your ass handed to you by these players in the same way? I think, to some extent, yes. I think it’s wildly competitive, and I think that’s why I like—

Sorry, not to interrupt. Just to understand Harry’s story, then I want to hear yours: is that just because it’s a multi-stage fund, that even at 500 it’s a bet on a bigger outcome? Is that the story, or is it just froth?

I think it’s—

Tied to your multi-stage point.

It’s tied to the multi-stage point: they just want to deploy 100 and 200, and—

Yeah, and if it works out at 5 billion, then they’ve leaned in on the position. Is that the bet they’re making at 500?

Yeah.

Yeah. So, how do you compete? Is the question—just to frame it, just to understand it.

First of all, yes, I totally agree. It’s incredibly competitive. So, let’s even take that comment on their motivations for a second. There are a lot of different things, and it really is the same Rob Go essay, just extended up.

My big aha, stepping back, is that almost everything about a big fund is better for the entrepreneur. The only countervailing trend is that if, in the end, a big fund doesn’t make acceptable returns, then the LPs at some point will withdraw the money from the big fund. But until that happens, almost everything about a big fund is good for the entrepreneur.

Let me tell you what I mean. With 1 big fund, you do more deals. You have 30 deals and more news. It’s more exciting to be in a portfolio with 30 names than 10, right? You have more news flow and more good things happen, because even if you’re just a freaking index, the reality is that if you do 30 deals a year and we all end up with roughly the same picking ratio—and we saw that analysis, Benchmark versus Andreessen—the picking ratio is less, but the volume is more. You end up with a constant flow of good news.

My big aha has been this: walls of capital give a lot of advantages to those funds in terms of winning deals, because they can pay more. They have the implied, “We’ll stuff more money in later.” It’s all an option. They can pay more because they just have more money, and at some level they may as well want to use it up.

The arguments against are very second-order. You’re saying, “Hey, but this junior partner at this big firm might not be there.” And the truth is, someone who’s 24 doesn’t factor that in, right?

These walls of capital are quite a powerful force, and I give credit to it. I think 2 things will be true 10 years from now, but it actually won’t matter a damn to me. Either, not only are they powerful, but they’re also profitable, in which case that’s where the industry will go; or the second outcome is that they’re powerful but not as profitable, in which case that capital wall will recede over a long period of time.

But both of those are well outside the event horizon of me making money in the next 5 years.

So, back to the thing, I'm just like Rob Go—one-stage VC. You've got to figure out a strategy in the context where the obvious consensus bet is going to be done at a price that you probably can't afford, and done by a fund that just has more to deploy than you because, using $10 billion of LP dollars, they bought deals up and down the stack. That's where you're playing right now, and you have to be better across the board. You have to—

It is worth it to buy deals. You're right. I mean, it seems silly. You shouldn't deploy your whole $10 billion fund for deals, but deploying $500 million of it to keep in the flow—it's like doing a $100,000 check in a hot AI company. It's the same. It's the exact same thing.

Yeah, I'm going to defend it. It's even worse—maybe better. Even if you're not doing it to avoid news, it's not a core strategy. Just by having that much money, you make news. I mean, I think Marc said, "We want to be powerful," right?

If you have a shit ton of money in America, you're powerful, right? Powerful is not the same as good, but it's a pretty big help to be good, you know? Right? I mean, it's the Russian—

It has to be at the top of every list. Every PR does matter, even today.

Remember, even though I've been wrestling with this, even talking to LPs, they're like, "Will this model work?" And I'm like, "Let me be clear: the only person who has an incentive to figure out if this model will work is you, Mr. LP investor. The entrepreneur—if I'm an entrepreneur and some big fund is willing to give me $50 million at a $250 million valuation, right?—I don't care that the 7 other deals they did don't work or have subpar returns. Did I mention I got $50 million at a $250 million valuation? I'm happy."

The entrepreneur doesn't want to call timeout here. The ultimate fate of this market will be predicated on how much money those funds make, and I'm not going to speculate on that.

I'm so sorry. I just disagree with you completely. Fundamentally, entrepreneurs do care if you have 30 companies, because they know that they're not going to get much time from you. And Rory, they really value your wisdom. They do, but they do. No, they do.

Absolutely. Twitter. It's weird. It's crazy.

It's not. It's not, because I get the emails and tweets about you. So it's not, number 1. And then, number 2, they do care that you've got a load of—

They do care that you've got a load of dogs, because then they're going, "Will he still be there? His portfolio's shit. And if he's not there, am I going to get someone else who's crap?"

And you know what? I've heard horror stories about that shit that turns up to my board, who never did the investment, and I don't like him. Oh, God. And then I also have signaling risk as well, because now I've not only got a dog on my board—not Rory—I've got the real risk of, are they going to write another check?

All those things are factors, but I'm just saying the competition from that money is real and meaningful. My bigger point is this: look what we said in 2010. We were happily doing SaaS; there were a bunch of people doing crypto, a bunch of people doing consumer, a bunch of blah. It was all spread out.

Now the consensus bet is enterprise AI. The wall of money is coming in here, so once something emerges and once the numbers make it obvious that something's working, pricing just goes—

Yeah, pricing is going to go to the point where you're pricing in that 2x best case, right? So you have to get there before it's self-evident. It's just that simple. You either have to have markets that aren't obvious or companies before they fully emerge, right?

Rory, do you think—I'm so sorry to be personal and blunt on a show—but you're doing enterprise in Silicon Valley at Series A. Do you think you're doing that?

When we succeed? Yes. Well, we do. And you're right: some of them you kind of go, "Oh, people haven't bought into this market." It's a case-by-case basis, but you're exactly right. There are a large number of them where you go, "Okay, I figured it out about the same time as everyone else," and it's a 10-term-sheet shitshow. You're exactly right.

No, your win rate goes—I mean, I look at our win rate. It obviously has gone down from where it would have been over the last 10 years, just because that's what happens, right?

What would you say your win rate is?

You know, I would say 50% to 60%ish, down from before. I mean, it's a little like win rate when you ask salespeople, "What's your win rate?" What really happens is they filter out all the shit that's not going to close. You can tell the vibes when it's not going to close.

So, to some extent, I think it's down probably 20% or 30% from where it would have been, probably 5 years ago, right? And you just feel that you're up against every excellent firm. So, yes, sometimes you win and quite a lot of times you lose, right? And you've got to be sure that—

So, therefore, if you find something just at the cusp of product-market fit, before it's obvious, right? Then your win rate can go right up and you can get a much better price. The market for consensus is fully priced in and fully discovered.

Yeah, your other point—which I didn't fully reflect on until you made it, and it's obvious—is the democratization of venture makes it stressful. Now, in every category that breaks out, everyone's competing, right? Again, there used to be crypto specialists; there used to be everything. It's good, but it does add to the stress when, instead of competing with a subset of venture, you're competing with everyone in the industry. Good for founders, probably, but it adds to the stress, to Harry's point.

Given the discussion that we're having, in the next 3 years or 5 years, will we have fewer, more, or the same number of seed firms that we have today?

You don't need to think. The answer is fewer. Half of everybody's got to die.

You're also seeing—you're seeing—

I think more. I think more.

Getting more.

Yeah, because—and I don't disagree with you here—here's just where I see it. Okay, listen. First of all, there are a bunch—I mean, raising a seed fund in 2020–2021 made no sense. Literally, folks that had never closed, had never had a markup, let alone an exit, were able to raise an 8-figure fund. That will never happen again. That was a weird time.

I have folks that were—it's happening today, but where it happens today, they're folks that quickly got into hot deals. So those funds are all going to die. Folks that raised at 0.1x lifetime or 0x, they're all going to die, right?

The challenge, I would just say—and listen, all the LPs I have, which are fewer than yours but some overlap with Harry, they're all doing very few new emerging managers. So if you view it myopically, you can say that's the combination of the 2 means it's got to be half, if you're being generous, right? But I don't know, man. There's so much greed, there's so much money to be made today, that if you have a hot hand in venture as a manager, you're going to get the LPs to give it to you. So I just think there's so much money, but—

Jason, yeah—

Those 2 sentences aren't contradictory. I think your comment about having a hot hand is spot on. If you have a hot hand, you're going to get a lot of money. But the truth is, in a world of fewer, bigger winners, fewer people, definitionally, have a hot hand.

But I don't think so, because at bigger firms, 8 people get credit for the deal. And the guy that actually did it at a big VC and gets brought into carry, that guy's going to get his own fund. So there's an endless supply: for every great deal we have, I think there's a new seed fund. For every great exit we have, right? And there are more of these big exits, right? And—

Jason, can I give you shit? For every great deal, you have a new seed fund, but as you correctly pointed out half an hour ago, there are only a small number of great deals. This is the real insight: if you can go public at $50 million, 200 companies are going to go public a year, which is what happened in 1999.

If you can go public at $150 million, 100 companies are going to go public, which is what happened in 2020. And if you can only go public at $350 million, maybe 30 companies are going to go public.

Yep. That's where we're at. There's going to be fewer winners. And the one thing, in the end, you need to stay in business is winners. There is less of the thing that gets people money.

The people who have those winners will have bigger winners and will get more of them. They'll get more money as a result, but there will be fewer of them. I mean, no, I can't argue with the math.

I guess my meta point—and, Harry, you're the boss; we can move on—is that there are a lot of things that, especially folks who've been in venture for a while, want to say with a lot of thought: "YC is overpriced," and blah, blah, blah, blah, blah. But I just think that the idea that seed is going to really contract—I think we're conflating the fact that there are a lot of crappy seed funds.

I don't think they're the same. I think we're going to see a thousand flowers bloom from every decacorn.

To be clear, you could argue the number of seed funds and the amount of seed dollars go up by 20%. The interest in the category is still there. The signal of success is more concentrated, so the people who have concentrated success are going to be able to raise more.

I think you could have half the number of firms and 25% more capital because anyone who has success will have bigger success and will raise more money.

Well, look, let me make one last point. I hear you. I have 1 investment that has not IPOed, and it is, I would say, only reasonably hot but very good.

That investment, where I was a seed investor, has already spawned 2 9-figure funds out of it. 2. And this one hasn't—this isn't Figma yet. It's already spawned 2.

So, if my one—this one unicorn—has spawned 2 9-figure funds, both of whom have raised 2 funds out of this, so more than low 9 figures, if that is not a big deal, I'm not that great. That's not the one time it's ever happened in venture.

If that's happening across these hot deals, it's going to happen repeatedly. We're going to have a ton of seed funds, right? Even if all the other ones should die. This idea that, oh, woe is me, I can't raise a 3rd fund and I've never returned any capital—tough fucking luck, right?

But just look at that story: this company is years from an IPO, and it's created 2 new funds, both with multiple funds and 9 figures. That's a big deal, isn't it? Maybe I'm wrong, guys. You can both invest in 1 seed firm today. Which seed firm do you invest in?

Jason's, of course.

No, no, no—mine or Jason's. Very kind. I love the way you went to Jason's and not mine. That's very kind of you, Rory.

We work with lots of great seed firms because we're doing the Series A. Let me, Harry—there are many things. Let me finish. There are many things I'm prepared to contribute to this podcast. What I am not prepared to contribute to this podcast is putting a gun against my head and blowing up my business model.

Just like my children, I love them all equally. I love all of those equally. You're not getting me on record on this one, big guy.

Jason, if Rory's not going to, as he said, delightfully shoot himself in the head or whatever it was, what would yours be?

Harry, honestly, I want to answer your question. If it was even 10 years ago, I would have always had an answer, and I would always tell the LPs, right? I'm sure you're asked too, and it's a great way they get deal flow, right? Who would you recommend, Harry?

I don't know. I just see too many folks who, when you squint, are chasing the deals. When you squint, I've seen too many folks who, when I actually see the numbers, aren't what I would have expected, right?

I'm sorry. I am a little hesitant to answer your question because I don't actually feel competent to know. Ironically, I recommended you in the early days, Harry. I've recommended some managers. I got off the ground when Christoph Janz from Point Nine recommended me.

We've all benefited from that flow, and I've recommended many managers. But if I was asked today by my anchor LPs, I can't answer the question. Maybe it's a failing of mine, right? But I can't answer the question because too many things are not what I thought they were from the outside.

I would give the same answer but from a more positive spin, right? In a sense, rather than you're knocking it. One of the things we have done is we've looked—look, you want to partner with seed firms. So, you say to yourself, who has the deals you value, and is there someone that over-indexes?

About 5 years ago, we did this exercise. We literally took every seed firm, and we have our internal CRM of deals that we rank as hot or high—in other words, marquee deals that we're really interested in. We ran it against every seed firm.

The big aha was this: it's not like there are 2 firms that jump out at you and have most of the great deals you're interested in. It's so distributed, right? And the aha is, we ended up with a Goldilocks list of 50 or so firms.

In each case, only—I think the highest one was only 6% of the deals that they were in—were things that we thought were interesting in enterprise software. So, it's a small number of deals that everyone has that are good, right? And that's just the nature of the seed business.

The genuine answer to you, Harry, is this: you have people you enjoy working with. You have people you just connect with personally. But the objective facts are, there are probably 40 or 50 seed firms where we go, “Wow, they're really good. They have their shit together.”

Which of them gets into the good deals? Hell, who knows? You just have to stay on top of it, right? I mean, that's, in many respects, what it gets back to. This is a scarily efficient, competitive business.

And the idea, Jason, that 20 years ago it was like, “Oh, those 2 or 3 people are great”—now you're deluding yourself. There's a couple of hundred type-A-driven people competing for vast amounts of money here. You're not going to find that you're the only good soccer player on the field.

Sorry, Harry. It ain't like that anymore. It's who's in the specific deal that you want at the specific time.

I'm going to ask 1 final question before we do a quick fire. We just have to—the cultural side, guys. We promise we never go into politics, okay? And I'm sticking to that.

We don't, but there's 1 I'm going to ask.

Coldplay concert. Astronomer. Just 1 question, Rory, before you jump on me: is Astronomer the company better or worse post what happened? Is all PR good PR?

I think they're worse off in the sense that they have to do a CEO search, which is always tricky and has a 1-in-3 chance of failing. So, that's a real momentum hit, and I've got to believe internally it doesn't feel great.

I'm going to even say something here that I saw someone else say on Twitter, and I'm like—and even the 2 people involved and the families of the 2 people involved, there's been an amount of schadenfreude. It's quite a sad thing, really.

You start with the humor, and then you quickly progress to the sad. And then, if you're on the board of the company, you progress beyond the sad to, “What a monstrous pain in the ass.” I think you had no choice but to make the change, actually.

Someone asked me, if you were on the board, would you let that person go? I think the answer I gave reflects the question you just asked. I think you had to—not just because of any policy issues, but because, as a company, you have to move on beyond it.

If you don't, yes, you'll get lots of meetings. But if you left the person in place, then every meeting he took for the next year would have an overlay. There would be a subtext in that meeting.

So, I think the company—it didn't help him. Yes, it gave him some notoriety, but it's kind of a frictional drag. The only way they could put it behind them was a CEO replacement, and that's a 1-in-3 chance of failure. So, net-net, slightly worse off.

I remember when this happened to Tiger Woods. Remember when the “Go On, Be a Tiger” program with Accenture launched literally the week after his extramarital affair came out? There was about a day when they thought, should they pull it or not?

Then I think everyone went home to their spouse, and their spouse looked them in the eye and said, “There's not a decision here. You're pulling that ad, dude.” And it was gone the next day.

Everyone recognizes personal tragedy. No one wants to, so it's just too shitty to touch. Move on. It's sad. Move on. Let them pick their lives up. Everyone makes a mistake. Go do something else, but the company needs to just put it behind them quickly and clearly.

Okay, we're going to do our Koshi quickfire. Koshi is a performance marketplace. They do the bets and realtime events.

Number 1: will the US tariff rate on Canada be at least 35% on August 1, 2025? Yes returns you 2.14. No returns you 1.44. Who knows? It's a random number. It depends on what the episode of the reality show requires.

It's 35%. I'm going to go with no, but it's an uninformed opinion. But then, frankly, everyone involved in this issue is uninformed. Why should I be the only one who isn't?

I think we're just all at a phase culturally across borders where we're digging in now. Harvard's digging in. Everyone's digging in. So, I say it's still there.

Interesting. Cool. Good thought passes.

Will OpenAI release a web browser this year?

Yes only gets you 1.25. No gets you.

Yeah, I looked at the odds. I'm cutting you off only because I'm actually interrupting you, but I'm the one who told you to read the odds. Yes, I think of course they will.

No, I mean, it’s hard to imagine the most ambitious company out there not doing something that Perplexity and other people are doing. Yeah, it might have to be yes.

Well, listen, the internet sure seems to think the answer is yes, right? I just think they don’t have to rush something here. They can be a fast follower. So, it seems 100% that it’s going to happen.

But this is a math bet, right? I’ll take the no just financially, but I believe it will happen. If this was a gentleman’s 50/50, I’m going to go for the money and just say that it might go back in the oven.

I’m with Jason. I think it’s GPT-5, agent mode, and shopping prioritization. I just don’t know if it’ll be this year. It’s August already.

Okay, good bet.

Final one: Will xAI release a Grok macOS app before the end of the year? Do you guys use Grok?

Yes, as I said, whenever I want to talk like Harry, I talk to Grok. So, I genuinely do use Grok. I test it. We’re all a little bit Twitter addicts. We need to get over our addiction, but it’s hard. So, I do use Grok as part of that. It’s not my go-to.

Well, first, going back to last week, I think Grok is wildly underestimated. And I think when Elon says something, he’s either upset and I get upset. I get upset. I even get locked in, it turns out, when I’m coding, when I’m vibe coding. But when he’s not upset and he says something, just like Sam Altman, you should listen, right?

So, when he said a couple of weeks ago, “You don’t need to use Claude Code. Just take all your code and stick it in Grok. This is what we do, and we find all our bugs and push to production,” he’s saying, “This is developer-ready.”

Interesting.

Okay, that’s what he’s saying to the market. Now, have they built an IDE? Have they built out all the pieces yet? No. No one on the thread I saw said, “Hey, Grok is not good enough for that.” They just said, “We just don’t want to throw all our code into a context window.” But that says to me this is real and on the list.

Having a macOS app is the easy version of that. You’ve got to do that, right? You’ve got to have a desktop app. Given that the odds are unknown, I’ll put a grand on this one.

Well done, boys. We covered a lot of ground in that. I so appreciate it, Jason. Thank you for putting your neck on the line with some of that stuff.

I mean, genuinely, I want to thank you very much.