Navan申请上市、Canva踩下刹车：原因是什么，接下来会发生什么？

Here's what nobody wants to admit: when LLMs finally work at something, the implementation will be boring as fuck. Harvey isn't some breakthrough in legal AI. It's ChatGPT with a law costume. Lovable isn't revolutionizing code. It's Claude with pretty buttons.

Price is the lever for 90% of economic transactions. If you can get a stupid price in the public market that's higher than the stupid price you're getting in the private market, then at the margin, most of you should go. With all these people leaving, is there any loyalty left in Silicon Valley?

This does bother me. Everyone who's getting $1 billion was in the room when the magic happened. The ability of the US investment banking business to shovel out the door is unparalleled.

And the truth is ready to go, [Music] guys. My favorite time of the week. What a roster of content we have this week.

I was looking at where we were going to start, and Rory and I were texting, saying that this was a good place to start when it was coming out. It was about Daniel Gross and Nat Friedman potentially being acquired by Facebook and/or Meta. I wanted to start there. How did we think about this following the Scale AI acquisition, and how did you guys analyze it?

Well, I think there's a lot in it, and we should spend a little time on it. My mental model, zooming out, is: why is Facebook doing this? Why are they buying? The second question is, why are these people selling, and what does that say? The third—and I think also very interesting—question is, why is it going down this way? In other words, why are people able to extract this kind of value for their labor, and how does, for example, California—and us being a non-compete state—impact that? So, there are lots of things to unpack, but maybe let's start with the first one: why is it existential to Facebook?

The only logical thing you can be afraid of is, as some of the writing has said, some kind of meta-model—no pun intended—that basically becomes your primary interaction with the internet, with the web, and has all memory about you, which is obviously where ChatGPT is going. It basically sucks attention minutes away from Facebook. In other words, if they don't build something like that—so I'm interacting with Facebook, I'm interacting with news, and I'm also interacting with whatever their version of ChatGPT is—the fear is that precious minutes of attention, which means precious minutes of money, go to ChatGPT.

Maybe that's obvious to all the listeners, but it's just worth stepping back and saying that's what it is. It's not that they want to open-source Llama and make money off it. They won't. It's not that they want to even have an API offering of Llama, something like an Anthropic offering. That just won't be big. The only thing that makes sense here is if you think that the minutes that people spend on Facebook will become minutes they spend on ChatGPT, and you can't let that happen as Facebook.

Once you perceive it like that, you go, "Maybe." If that's going to happen, you've got to do something about it.

I'll tell you my rough guess, just thinking through it a little bit more. Meta, as we do this, has a $1.8 trillion market cap. My guess is fairly simple: Zuck's put $100 billion to this—$100 billion to catching up and maintaining dominance—and that's the budget. The budget probably can't be $1.8 trillion. That would be high. I mean, it is possible to sustain that dilution; there are deals like that. But in all seriousness, when you look at Scale, when you look at trying to buy everybody for $20 billion, right, and Perplexity, it kind of ties to having a $100 billion quick M&A budget to get back on track. It's 8%, I think, if I'm doing my math right. It doesn't seem outrageous when you think about spending a quick $100 billion—8% to get back on track.

And just to say on that, I can simultaneously believe it's totally a bad idea and it won't work, and it's totally a good idea to do it just in case it might, which is the really zany thing about it. I didn't buy into Oculus, and I'm definitely not as convinced that you need to spend this money or that the way you're spending it will be successful, but I totally get it.

When you're the CEO of a $1 trillion company, there's a limited number of buttons to press, and you want to press a button. The only kind of button that a CEO of a $1 trillion company presses is a big button. You don't go to Mark and say, "We've got an existential risk. Let's spend $2 million," because that's just not what CEOs do. There are only a few buttons you can press at that level. There are only a few places you can buy this kind of talent, and there you go.

Just to comment on it, Sam Altman quote-tweeted Similarweb today. The App Store obviously isn't all of AI and all of the world, but it was interesting: over the last 28 days, ChatGPT had 29.5 million downloads. ChatGPT—29.5 million, just mobile. TikTok, Facebook, Instagram, and X had 32 million combined. So ChatGPT's mobile downloads are just about equal to all the social media guys combined.

Yes, exactly. I mean, the first rule of owning a $1.7 trillion business that spews off $100 billion a year is: don't blow the $100 billion. Don't lose it, don't kill the golden goose, and don't let it be killed.

Taking the conversation in another direction, I was thinking about what all these acquisitions have in common and what makes them interesting and different. If you think back to what happened here, everyone who's getting $1 billion was in the room when the magic happened. And that's the soundbite. Let me tell you what I mean by that.

A whole bunch of people tried to build these LLMs, and the early OpenAI team did it. Everyone who was in that room and knew how to do it went on to build some version of this kind of outcome. You either stayed at OpenAI, you peeled off and went to Anthropic, you peeled off and went to Safe Superintelligence, or you peeled off and went to Mira's new company.

No one—and for the record, no one in any of the other rooms where the magic didn't happen—is ever going to get that kind of money. I respect people like Cohere, I respect people like Adept, and the other ones like Inflection, but none of them have magic-moment money.

It's really interesting if you think about it. That kind of thing happens occasionally in history, in industrial history, where someone just figures something out that's so important that everyone who was in the room when it happened has value just because they know. From there, how does that shape out into money?

I'll give three examples, or maybe four. When the Chinese figured out how to make silk, they basically executed anyone who tried to tell anyone else. Problems dissolved. If you look at William Shockley and the early transistors, or the Bessemer steel process, when people tried to leave, they just litigated them out of it.

The great thing about California is that we're in a non-compete state. If this had happened in a state that allowed massive 5-year non-competes, all those guys would be sitting at home getting that $300,000-a-year salary, going, "I can do this in 2 years."

One of the amazing things about California, I think, is the fact that since the 1870s, it's been really hard—and in fact, it just got even harder—to enforce non-competes. All these people were able to leave, rely on the doctrine of inevitable disclosure, set up their new company, and, again, not copy the past, but know how to make the magic.

That's the aha here. It's just super interesting that if you were in that room, one of those 20 or 30 people in California, you can go away and effectively sell that knowledge that you have that no one else does.

Does that not lead to the commoditization of magic, then, in a world where there's some relative commoditization?

You're exactly right. It would be better if there were only one, but it doesn't lead to everybody having it, because it turns out the price of buying it is a couple of billion bucks, right? But, yes, there's no doubt there is leeching of knowledge out. And as I say, if you look at all those examples—Shockley and the transistor, the Bessemer steel process—everyone's always trying to stop the magic from getting out so they can extract monopoly profits.

But over time, it gets out. There are layers of talent acquisition value, which is your Scale with your $14 billion, and your Nat and Daniel, and your Miras. Sam was on Jack’s podcast, and he said that OpenAI’s talent has been offered several $100 million offers by Meta several times.

I guess that’s the next layer of talent, where maybe they haven’t seen the magic, but they’ve been in the building when the magic was there. I saw something very cynical that said—and if it is, it’s even more impressive—this could be total jiu-jitsu. If you’re not getting offered $100 million now, and Meta calls you when you’re down the hall at OpenAI and offers you a lousy $30 million, now you’re insulted, right? Maybe they’re just messing with their heads.

The other thing is, it’s going to be really hard to be the VP of HR in charge of the Meta LLM project. If you’re sitting there on your $2 million a year thinking you’re killing it, and then suddenly you discover the new guy is getting $50 million, your head’s going to hurt, right? I don’t know how much of that is just very clever disinformation, but at some level, you’re right. Intuitively, if you know how to cook this stuff, if you know how to make this magic, you have value. In California, it’s very hard to stop you from monetizing that value.

Harry, how much money would it take for you to dump all your 20VC LPs and your listeners and go join Meta?

Genuinely, I really wouldn’t. I wouldn’t know what to do. I hate working for someone else, and I get great discomfort from being in large companies. I have enough money now that I can do what I want. I’m really happy, for one of the first times in my life, honestly, without being sappy. I’m in a really good place. I would be miserable doing that.

I go, “Mark, just give me a number.” A billion. We’ll go running together.

Okay, done. Done. Email your LPs. Tell them, “Tough, tough, tough luck. They just gave you the latest $450 million, and now you’re showing up.” You do have to spend 4 days a week in the Meta office in Menlo Park, but there’s great running on the trail on the Peninsula. You haven’t done these runs. They’re great.

It’s a billion, invested over 5 years. We’ll just give your LPs back their money with a nice—spoiler alert—something that no one knows.

I was offered $75 million for the 20VC media company in the last year, of which I own 100%. I went for a walk with my mother and I said, “What should I do?” She said, “What would you do tomorrow if you sold?” I said, “I’d start the 19minute VC.”

She looked at me and said, “I don’t think you should sell if that’s what you choose to do the next day.”

She’s exactly right. Moms are always wrong in this. They give you this great advice from the heart and the soul, but they sometimes miss how the stitching works together in the venture industry, with the LPs and the holdbacks and whether you have to work for another 4 years. Moms are directionally correct in the heart, but sometimes they miss the details in these deals.

It depends on how much you have. The marginal utility of the first dollar versus the $75 million is very different, so a lot of it depends on your personal position.

That’s super clear. But also, to be fair, I mean, I would imagine in some cases part of the attraction has to be the ability to play, right?

Yeah, the ability to play. I don’t think it would be more fun running 20VC in a media empire for Rupert Murdoch. It might be more fun running an AI project for Meta when you’re literally told, “Spend $100 billion to make it happen.”

Whenever we sell companies and their CEOs go join a bigger company—and Jason can smile at this—I put a little note in my calendar to check in after about 3 months, because they’ll need some therapy, right? I remember one of the guys said to me, “I’ve accomplished my day at 9:07, and then the rest of my morning is just about not getting into trouble by saying stuff.” It’s just a very different gig. But that depends on what you need to do.

Is there any loyalty left in Silicon Valley?

I do. This does bother me. Where’s the loyalty to your LPs? To dumping your LPs? Where’s the loyalty? I can think of a lot of folks in tech that we look up to who dumped their LPs, who quit their unicorn to go into venture or other deals. I get the rationality of it. They quit their fund to go work for Meta.

At least Alexandr Wang from Scale gave his VCs back $15 billion. At least he did that. He did it right. But people just leave ship. There’s no—I find this a little gross about all of this. Maybe loyalty is dead.

It’s great that everyone at OpenAI who was a founder, except, I guess, Greg and Sam, is gone running their own competitors. But there’s also something about it. I don’t know. Maybe it’s Sam’s fault, but I don’t like it. I don’t like everyone dropping everything and leaving everything.

Now you’re sounding like Bill Shockley, or whoever it was when Fairchild happened. The Mr. Fairchild guy.

But if Alexandr Wang’s highest and best use is as a senior employee at Meta, then isn’t it wonderful that the free-market system was able to pay everyone else $15 billion? All power to Accel, which probably booked a $2 billion gain here. If that’s disloyalty, any of my CEOs who want to be disloyal to me and give me $2 billion, I’m in, baby.

I think the system worked. When lots of money goes through the system, stuff happens. Relationships become hyper-transactional, and maybe that’s okay. Maybe it’s okay. It just creates really interesting expectations between VCs and founders, and founders and management.

There’s very much a vibe of, “Take $5 million, $10 million, $15 million, or $20 million from my investors. Don’t work out? Goodbye. Here are the keys.” That’s not how I grew up. If I had grown up that way as a founder, my investors would have made nothing. If I could have left the keys on the table, there would have been 2 or 3 times when I would have just said, “Here you go to my VCs. Enjoy running my e-signature company.”

I hear you. It’s just different sometimes.

Years ago, someone sent me a note—an actual VC, one of the founders of Charles River—and said, “You always have to play the game by the current rules.” There was a time when capital was scarce, and you had an obligation to your investors because you weren’t going to get more. It was get-rich-slow anyway; SaaS is a compounding business.

The truth now is we’re in the exact opposite of that. We’re in a “you’ll know when you know” environment. I think a lot of these things, once it starts to hunt, you make a lot, and if it’s not hunting, you don’t. When people are making lots of money, the institutional glue gets a lot weaker. It’s just the nature of the beast. There’s no point getting frustrated about it. Just play the current game.

Guys, speaking of playing the current game, Harvey raised $300 million at a $5 billion valuation. This really stood out to me as a round. How did you analyze it? How did you break it down? There are different reports of where revenues are for them.

I don’t know what the revenues are, actually. What’s our best guess based on scuttlebutt? I think they’re at an end-of-year run rate of $100 million by the end of this year. I guess the growth rate’s more important, right? They’re approaching $50 million or something—$100 million ARR, growing not quite at Rippling rates, though.

No, but pretty fast. Hold on, let me check—it clearly says $30 million.

All right, he’s got his cheat sheet out. I love it.

To me, it’s super impressive in a way. I know a little bit about the space. I know a little bit about the legal needs, and I know about the legacy players. I never met the Harvey team, but I talked with a lot of founders doing similar things.

My problem was that all the apps were great. If you run a set of legal documents through OpenAI and then ask it to analyze the terms and conditions of a 300-page legal document, it’s great. If you ask it to research the current status of California’s auto-renewal law, it’s great. Every legal AI app that pitched me, I saw a demo and thought, “They’re all great. They’re all curing cancer.”

Kudos to the VCs, because if I’d met Harvey at the seed stage, I would have said, “This is great.” But I don’t know that I could have told the difference from the 11 other companies that were doing this amazingly, essentially as wrappers. I think they did some things really well.

There were a bunch of people doing it, but they started off by grabbing hold of OpenAI. They became the de facto winner in terms of Silicon Valley presence and lawyer perception, frankly long before the product was there. They established what I think of as intellectual mindshare as being the lawyers’ choice super early, when the product was still mediocre.

We looked at some other companies in the space and did references with a Harvey customer. The reference was some version of the following.

But my partnership said we needed to do something in AI. These guys had a big story. I gave them $1 million. It wasn't doing that much for me then, but I had to have an answer, and I had faith that they were on the right journey.

They marketed this thing as being limited to a certain number of customers. They made it scarce, signed some early customers—Allen & Overy and, I think, one of the accounting firms—and got this kind of stampede effect going. It was brilliant. I think the product frankly lagged that, but over the last couple of years, they have filled in behind it.

In classic Crossing the Chasm—or, more like Geoffrey Moore's analogy of the tornado—they claimed the space, followed up with the engineering, and now they have a compelling product. They claim to have 300 of the, I think, law 500 or law 10,000. But they did it early. They thought bigger than some of the other people in the space who thought it was just, “Knock down deal by deal, do good work, be earnest, make a good product.”

Those guys were like, “No, make noise, freeze the market, declare yourself the winner, details to follow.” And they pulled it off.

If you're doing this at $5 billion, what are you underwriting this to in an outcome scenario plan?

I mean, obviously, the model is going to give you the bland answer: at least a 3x with upside to 5, because that's the correct answer. But obviously, your real question is: can this be that kind of outcome? I think there's only—so let's talk about that.

Most legal software doesn't have outcomes anything like that, because most legal software—there are roughly 1 million lawyers, and we joke, but 1 million is actually not a lot of anything. A thousand bucks gives you a billion-dollar TAM. At $2,000, it's not a huge market. Traditionally, outside of litigation, it's not a huge market if you're selling software-like stuff.

Fun fact: there are 2 escalators of value. The first is this: for every dollar lawyers spend on software, they spend $5 on Westlaw or Thomson Reuters for actual legal information. Those used to be the old books you saw when you thought of lawyers in the 1960s. Now, obviously, it's online, but data and information is 5x the spend of software. That's the first argument.

The second argument for an even bigger TAM is, obviously, if you make some kind of AI lawyer, we can charge a lot more than Westlaw because we literally eat the work. We replace the lawyer, so you can get paid as if you're a lawyer.

All that is to say, you can't get the kind of outcome you need to make this work if you see it as just another piece of legal software. The math doesn't work. If you see it as an adjunct to your research tool, that gets you closer, but you probably literally have to believe it's doing some of the work to make the TAM math work.

I agree, and I think it goes back to a question that you said before, Rory, which I think was one of the best statements that we've said in the last few episodes: Are we going to see AI software providers be able to eat human labor budgets? If so, then we have the holy grail and we're all going to do very well. If not, then we're overpaying.

On the Harvey side, it was interesting hearing Kim's thoughts. I did meet a little while ago with a legal startup doing something very different but essentially tapping into the same buyers, right? They quickly got to $20 million in ARR on very little funding, and ROI was super high. What they were doing with AI had super-high ROI, but it didn't really do all that much AI.

The point was that lawyers—the IQ is probably the second highest behind engineers—but the sophistication of the purchase was not particularly high, right? It was to improve lawyer productivity. It worked, but the quality of the AI was limited.

So it's a question: when we look at Replit versus all these others versus Lovable, there's a lot of a quality war there. I don't know if it's possible to have a quality war in the Harvey space. I just don't know. I don't know if lawyers have enough time to switch between 11 tools, dig in, and see which one analyzed state-law conflicts between Georgia and Alabama properly. Maybe they do, but I don't think so.

I think you're right at a high level. For the case-law stuff, it's the same for everybody, so it is: Is it the best tool? I think part of the value proposition from someone like Harvey is they'll say, “We'll integrate with your internal information.”

So, picking just a sample lawyer, it's not just the case law that they bring to the table. Obviously, we RAG and crawl through all your internal stuff, and we can bring that to the table as well.

And that's so easy.

Agreed, but our AI has RAGged 20 million pieces of SaaS content. It took a day—it literally took 6 hours—to input 20 million, to RAG 20 million words of content.

But you haven't lived the dream of having sold that software to a law firm. If you look at the typical document-management software that these firms have, it's 20 years old. It's a company like iManage. There does appear to be a high propensity to stick with even mediocre software, so my guess is getting in the door here has value.

You're frankly ruthless about trying new software, Jason. You will dump yesterday for today, and you'll dump today for tomorrow in a heartbeat, and you're 1 person running your own business. If you're selling to a partnership where, remember, everyone's your boss and no one's your boss, you have 200 high-attitude, pain-in-the-ass lawyers. Imagine a VC firm like that. I can't imagine it.

You get them up and running on this system, and 2 years later there's a better product. You don't need the heartache, because it takes time. You talk to them about their use of Westlaw—they've been using that thing for 40 years—and they're like, “I'm dying with this Boolean search here.”

So I think there's a real institutional value. Getting in the door and locking in those customers has value. It's clearly going to be a valuable company. I think the question, to Harry's point earlier, the only question left on Harvey is market size: market size and market-composition shakeout.

There's Leya as well, which is doing phenomenally well. I mean, they're the European counterpart, but I did references on them. They're beating Harvey in a lot of cases—a lot of cases. And then we invested in Solve, which is a vertical application for patent creation, editing, and submission, I think, and Crosby, which is next on our list.

Crosby is a law firm backed by Sequoia.

By Sequoia, exactly—the ones who did Harvey.

I know you're moving on. I do think the meta-question for Harvey—because, let's be clear, Harvey is still a B2B application at the end of the day, right? I don't mean to roll back, Kim, but will there be enough $50 billion- to $100 billion-plus B2B companies in the age of AI to justify these investments? Enough? Maybe the math ties to it, right?

But there aren't enough $50 billion- to $100 billion-plus B2B public companies today to justify these deals.

I agree. If it is software—even, I mean, let's just go for a simple, humble 3x: $15 billion. If it's just software, it probably doesn't, and it trades at 7 times in the end. When things get normalized, it's a $2 billion thing. You don't get there on the TAM.

So you're right: if it's all just legal software, you don't get there on the TAM. If it's legal eating the work, then it's 2 orders of magnitude larger. So that's the question: does it eat the work? And then, do you get paid when it eats the work?

Which is actually something we didn't talk about when we last did this. There are 2 separate things. Does your software automate what people used to do? If it does, then at least value is being created. But then the second problem you have is, if there are 3 people competing to make the same kind of software, the law firm gets the value.

Excel doesn't charge $60,000 a year because it replaced an analyst. It charges $60 because that's what you get for Excel. So can they replace labor and keep the value? That's the question for Harvey.

The interesting thing, segueing to Crosby, is that, just for everyone, it's a company that, as you say, Sequoia—one of the successful backers of Harvey—also backed. That's a company effectively using AI to offer, quote, “a better, more efficient law firm.”

They are, as it were, using technology to eat their own work. Obviously, the value proposition is that they'll offer a better product to their customers while at the same time presumably being more efficient. That's the bet, at least. I don't know.

I've already done that.

Oh, good. Tell me more.

Well, all the legal work I do, both for the fund and for SaaS Inc., I run everything through my AI and Claude, and then I run it by my counsel to see if it's correct.

How many times is it not correct versus correct?

Never. It's always right. That's because the stuff I'm asking about isn't that hard if you've indexed the entire world's internet knowledge.

For example, I never read my LP agreements. They're like 1,000 pages long, right? But for the first time ever, I had an LP who wanted to transfer to a third party. I didn't know how it worked. I had 2 options: I could send it to my old counsel, who charges $3,000 an hour and would give me a grumpy answer in 3 weeks.

Then I'd say, “Well, can we get on a phone and talk about it?” And he wouldn't get on the phone.

Okay. I have new counsel now. Or I just threw it into Claude. Claude analyzed all the documents and gave me all the correct answers. I talked about different scenarios in which the LP would want to transfer, and I said, “Write this up in a short memo for me.” I shared it with my counsel.

A couple of days later, he read the documents and said, “That’s absolutely correct.” Again and again, across all the documents. What I need is a law firm that works at the pace of AI today. I’m not working at the 2021 pace, where we worked 18 hours a week and had 3 jobs. I need my legal answer in seconds, and then my law firm can confirm it afterward.

I’m at the bleeding edge, but I love that these are the lawyers I have today. I’ve rebooted my legal team for people who can review my AI answers rather than the other way around.

So I buy that, right? There are a whole bunch of things, like vendor-management contracts and NDAs, where review should be 90% AI and 10% human check if there are exceptions. Therefore, it can be instantaneous unless there’s an exception, and you’ll be damned if you’re paying $1,000 for it.

I totally think that’s a thing, right? I think there are law firms—we’ve seen not just Crosby, but other firms—specializing in NDA review and very typical documents in a way that, interestingly, wasn’t doable, to state the obvious, 10 years ago. We looked at LawGeex 10 years ago; they were way ahead of their time, but the technology didn’t support it, so they weren’t able to build a compelling business at the time. Today, NDAs should be, at the margin, free.

The interesting question is how that manifests itself. If all the existing law firms are dumb enough not to get with the program, they’ll lose Jason’s business because he’s going to say, “I want the flat-fee, $50 review and a turnaround time of 15 minutes.” If those firms resist that, then you’re right: maybe Jay will do it himself, but typically firms like Crosby will be wildly successful because they’ll take the business away from the older law firms.

I’ve got to believe, though, that enough of them are just going to get with the program because, remember, they all now, in our new world, already have Harvey. They’ve got the guns, too. They’ve got the tools. It’ll be interesting to see, fast-forward 5 years, where the incumbents have Harvey or Legora, but have business-model inertia. All the new Crosbys are competing for Jason’s business and saying, “Dude, it’s $10 for an NDA, $20 for a vendor agreement, and we only charge real money when you have to do a complex transfer.”

My gut is that some of the old guys will be slow, but enough of them will adapt. I don’t know. It’ll be interesting to see how much the new guys can build in this space.

This is exactly my point, though. Venture is defined by 2 types of outcomes: 1 where you drastically underestimate the size of the market and it’s so much bigger than you thought it could be, or 2 when you overestimate the market size and see it fragmented and unbundled into so many different composite parts that it’s actually not as valuable as you thought it was.

I think that’s exactly the case here. When you look at it—when I was doing the diligence for Solve’s patent-creation business—there were 15 competitors. Patent creation, and every different legal-adjacent area, is the same. I think it’s as fragmented and unbundled as it is, which will actually make it a smaller market than people give it credit for.

I think it will. I do think there are about 5 or 6 legal, process-specific processes—patents and immigration, for example—that stand on their own. Then I think there will be the general corporate solution for general corporate law, which would be Harvey and Legora.

But yes, even peeling off 50,000 patent lawyers is significant. You peel off another 50,000 immigration lawyers, and then even up to 50,000 personal-injury lawyers, of whom 40,000 live in Texas, and pretty soon that trillion-dollar legal market is down to a half-trillion-dollar core of corporate litigators and contract writers. So you’re right: you start unbundling it. I mean, we did the math on lawyers.

The truth is, if you’re doing plaintiff law, you don’t need Harvey. If you’re doing patents, you don’t need Harvey. But if you’re Wilson Sonsini, you need Harvey. If you’re Latham, you need Harvey. If you’re a mid-tier law firm in Phoenix and you’re a 400-person local firm, you need it. You probably have the market, which is why the math only works if you start to eat that work, baby.

Otherwise, you know what the triggering thing is about the Crosby thing? This is the triggering thing. The truth is, we’re still learning, right? But what things like Crosby show is that it exposes how much human labor is going to be replaced by AI.

It highlights the mediocre with a blinding spotlight because, if that associate—forget about the senior partner; I’m willing to pay the senior partner whatever the fee is—takes a week to get back to me and it’s wrong, while Claude told me the exact answer from my LPA, those folks are all going to be out of a job. The mediocre—the mediocre in everything—there’s no need.

It’s not just NDAs. Claude alone can review very detailed commercial agreements. “Okay, I want to get out of my Salesforce contract. It’s 1,000 pages long. What are my options?” Claude can give you that answer in 5 minutes.

“Yes, it’s just a big no. You cannot get out of your Salesforce contract. It’s the rules.”

“But what if they sue me? What are the odds they’ll sue me? What happens if they sue me?”

Joking aside, you’re exactly right. AI is just going to pound on efficiency. I remember 30 years ago someone saying, “The internet ground out commercial inefficiency. It ground out middlemen like travel agents. Anyone who’s connecting buyers and sellers and that was their only thing—the internet exposed that.”

This is going to grind down knowledge-work mediocrity: anyone who’s just recycling stuff that’s easily known and is slow and unresponsive. You’re exactly right. It’s pretty impressive, and that’s the way capitalism works, Rory.

When you ask me next time why I’m responding at 1:30 a.m. within a minute’s notice, I’ll remind you that, to win, you either have to be the best or hyper-responsive. And I am the latter.

Yes, and we know you can’t do the former, so you’ve got no choice, baby. Listen, humility is crucial to everything. They both work. They both work.

I mean, venture—if you think venture is a service business, which most people think it either is between 10% and 99% a service business—it holds true. There’s someone in your investor syndicate who speaks from the top of Mount Sinai and knows everything, and then there’s the one who responds to you in 60 seconds. I don’t care about anyone else on my cap table.

Totally. Yes, guys, there’s so much doom and gloom every week. We also have new IPOs this week. Navan filed for its IPO. This is going to be a big one. I think that last valuation, Jason, you’ll be able to tell me, was in the $10 billion range. There will be several venture firms here who make a lot of money from this. How did you guys analyze this one going out and filing now, at this time?

IPOs by number are up 62.5% this year, apparently, based on data I saw today, right? Just about every IPO is up, and everyone’s ready to go today, right? A couple of weeks ago, we weren’t all ready to go. Everyone other than Canva, doing its secondary at $40 billion or whatever, or Stripe, is now planning its IPO. They’re just planning it today. This is too good a start to the year. Why wouldn’t they?

Yeah, anyone who’s got the numbers is going to go public in the next 12 months. They’re all going to go public.

Is there a limit to how much the public markets can take so quickly, with the stampede that’s coming on the supply side? Is there a limit to how much the demand side can ingest?

You know, Jason reminded me a while back that in 2021 there was an IPO a day. I think the ability of the U.S. investment-banking business to shovel shit out the door is unparalleled.

The truth is, when stock—look, when someone does Circle at the IPO, we’re all very Pavlovian, right? When you buy at $31 and 3 weeks later it’s trading at 7 times that amount, let me tell you what your little Pavlovian reptile brain says: “Do more of that.” The next one won’t be quite as good, and the one after that won’t be quite as good again. But as long as it still feels good, those rats will keep pressing the button.

No, there’s not a practical cash limit. As long as this stuff keeps working and nothing exogenous happens, they’ll be able to get deals done.

Navan is clearly a top 1% startup, right? There’s no question the metrics will prove it out. But even, say, being objective, is it better than Ramp? I know it’s not a direct competitor.

My meta point is just this: I only have one question. When you have a 0.1% company, but you're not sure it is the generational company, how do you know when to push all the chips in? To Harry's point, it's fun to do it, right? I've done it twice, and I don't know—it was kind of not to that level. It's fun to push all the chips in, but you've got to make sure it's at the edge of generational, don't you?

That's the Roblox point that Altos always makes, right? You've got to wait for that Roblox and then push all your chips in. Is Navan as good as Roblox? I mean, it's generational, but it's great. Is it as good as Roblox, though?

I mean, the fundamental point you're saying is, yeah, concentration without absolute excellence will beget subpar returns. You've got to be right relative to price, because the trick is not just that you've got to concentrate, but you've got to concentrate while the price is still attractive.

Remember, the other thing—it's just worth stating—is that on the last 5 or 6 IPOs, there's been at least 1 round on the private side that was clearly priced wrong. We've seen a bunch of them. I mean, obviously, we've seen it on Chime; you've seen it on Hinge Health. I doubt there was a Circle round that was priced as high as it's currently trading, so everyone is golden there. But in general, it's hard to concentrate, and it's even harder to concentrate and get the price right. So, yeah, when you do it and pull it off, you obviously get a stellar return, and more power to him.

Circle today is a $68 billion business, which makes it more valuable than Robinhood and Nubank, just to set some parameters, and even more compellingly, more valuable than Coinbase, to whom it gives half of its gross revenue. Will it survive falling interest rates? And is this peak meme stock?

Too extreme on both sides. Of course, it will survive falling interest rates. It won't go bankrupt. But there's no doubt that its current model is all about, as someone succinctly described it, letting people give you their money and getting to keep the interest, right? So, if the interest is less, they keep less. It's a far worse business at a 2% money market fund than a 4%–5% money market fund. So, to the extent rates go down, obviously it won't be as compelling. But it will survive.

And then the second thing is, is it peak meme? I think there's definitely some element of right deal, right time with all the crypto reform in the House and Senate. It just feels perfectly on point. There's probably a fairly thin-ish, probably reasonable float, actually, because there was some secondary, but it's the kind of stock that can run. It's an N of 1. The story makes sense and, yeah, it's obviously gotten carried away.

I don't for a second think anyone who has done any kind of math analysis thinks it's worth 57 times run-rate revenue. I don't think anyone thinks that.

But here's the question I struggle with. Kim, you've got more experience here than me. Forget it—we could talk about whether the IPO is mispriced, and someone could play the role of grouchy Bill Gurley, who's definitely smarter and more successful than me. Totally. But the last 5 days, it's up 46.4%. What changed in the last 5 days? Forget about making 5 times. Harry and I could have just taken our funds, put them all in 5 days ago, and made 46%. What's the IRR? I can't do the IRR math. If we did 46.9% in 5 days, what's the IRR annualized for that deal? It's awesome.

No, you're exactly right. There's no, “I'm going to wait for my portfolio to appreciate.”

Yeah, no, there's—look, there's no logic to it. I mean, you just have to say to yourself, these kinds of high—there's no logic to it. It's a trading asset, and it's all the symptoms of what you see in speculative bubble behavior: vast price movements in short periods of time for no information.

The efficient-market-hypothesis people get all mad and say there are no bubbles, and you'll go, “This looks pretty bubbly to me.” I think large numbers of small-volume, ill-informed traders in a stock where, relative to the float, there's not a lot of it—and you have no doubt in my mind that when the other 80% of the stock comes off in 6 months, I don't think it will be trading at 57 times revenues.

If you're looking at this as the Canva executive team, can you genuinely help me understand why you delay an IPO? You're looking at Circle being priced, but everyone else is enjoying the fruits of the public markets treating them well. Why do you delay?

It's a great question, because it's actually the only reason to pose the question. Let me tell you what I mean by that. When you have all these things—should you stay or should you go, to coin a phrase, on the IPO—and all the private is great, the real question is: Is the cost of capital cheaper in the public markets than in the private markets?

For a long time, the private market has been cheaper. We've been a wonderful source of capital. We give you money, provided you get a preference. We leave you alone. We don't bug you. You don't have to do analyst day. We're not really that mean. I mean, some people think we're mean, but compared to the guys in New York who run hedge funds and various kinds of funds like that—activist funds, that's the word I couldn't think of—yeah, we're nice.

So, it's been really nice being private. The only thing that's going to change that is not some kind of—I was thinking about it because you'd asked the question 2 or 3 weeks back: What would you change in the public markets? And I didn't have a good answer. You know why? I live—it's kind of the wrong question. Price is the lever for 90% of economic transactions.

If you can get a stupid price in the public market that's higher than the stupid price you're getting in the private markets, then at the margin most of you should go, right? And I think you're right. Maybe not—maybe people have ideological reasons not to. But let me tell you, if you're owning a Bitcoin trading operation or a stablecoin operation, you are typing as fast as your little fingers will let you. And there are bankers locked in rooms as we speak doing that, because price is how the public market sends a signal to the private market: “Hey, come on in.”

Would you not argue that's a relatively naive way to think about going public, based on a transitory moment in time of what public markets will price you at? I was at a dinner with a $10 billion public CEO last night who's a friend of mine, and he was just moaning about being public. Very simply, it's a transitory moment in time. You will appreciate and depreciate. Who gives a damn what you went out at?

Look, first of all, I'd say it's less naive than reductionist. And look, you're right: there are all the other negatives and positives of being public that are hard to change. My point was, at the margin, if you get all the grief of being public and, on top of that, you get a lower price than the private side—and a consistently lower price—then you never bother.

But if you get all the grief of being public and it's still a pain in the ass, but in return you get a liquid stock and a 50%–80% consistently higher price, then the argument is hard to resist. Now, you're right: if it's a flash in the pan and it's gone in 2 months, then, yeah, it would be a naive reason to go public. But if, on average, the capital is cheaper in the public markets, then over time it'll pan out.

But if we just go back to Canva, though, I just don't get it. It's a very strong consumer brand, very well known, with incredibly strong financials that we know of. It's north of $3 billion in ARR. Why would it not go out? They don't have to sell securities. They're kicking off cash, which means that they're net buyers of their securities—employees, not sellers. So they don't need to raise money.

I mean, let me give you the data-center guys: Backblaze, CoreWeave—you go public. I mean, fundamentally, you go public to raise capital.

What’s odd about some of these companies is that they’ve been private so long and have done so well that they’re past needing capital. They’re just kicking off cash, so it’s not an imperative.

If Canva generated enough cash, you could imagine it generating $1 billion of free cash flow or more a year—maybe $1.5 billion, right? It’s easy to see. They could have 40% free-cash-flow margins.

If you have enough secondary interest, too, you could just buy everybody out. There’s enough cash at $1.5 billion a year to buy out even all the growth investors—just about everybody—and no one listens to the guys who bought a few shares in the late rounds anymore.

But forget about that. Blackbird and everybody else could get as much liquidity as they want, right? Return the fund. If you generate enough cash, and the founders have given most of their shares away to charity, they’re not trying to buy the biggest yachts.

If you could generate $20 billion to $30 billion of free cash flow over the next decade, why go public? Maybe you don’t need to go public at all. Just buy everybody out and pay dividends as founders. Pay $1 billion a year in common-stock dividends if the founders own 80% of the common stock. Most of us could live on $600 million or $700 million a year in dividends, couldn’t we? I think even you could manage it, Jason.

And you’re right. Seriously, it’s not that they can’t buy it all out. I don’t think they could buy it all out—it would take too long—but I get you. I think they raised less than $1 billion.

Well, they might own 30%, in which case they need $10 billion, but it’s not crazy. Unlike a lot of these startups, when we talk about Canva, it is possible to buy them out, right?

Yeah, if you’ve raised little capital and have high cash-flow margins, you could buy them out. I don’t think that’s the reason, though. I think a lot of what they said is that they’re really trying to focus technically on the AI development because they have a lot of new stuff to build.

They just don’t need the grief and the distraction, and they don’t need to do it, which is a perfectly rational reason.

All I’m saying is, if the 3 of us were running Canva, and let’s say we were still growing north of 30% or 40%, like they are, and we sat around and said, “Listen, guys, we could buy out our last investors at 3x. It may take a few years to get there, but they’ll make the 3x.”

Let’s just chill. Let’s pay ourselves $1 billion a year in dividends, like the Basecamp guys do on steroids, and let’s buy out our investors at 3x when the time comes.

I don’t know a single public CEO who’s happy, to Harry’s point. I literally don’t know. Even the most successful ones—Palantir and Cloudflare are the 2 most successful public companies. They’re great, but they don’t seem happy, do they?

Do Matthew and Alex seem like the happiest people on planet Earth? They don’t. They’re driven. Mad respect, but I’m not commenting on Matthew.

Why? I would have that discussion with the 3 of us. Why don’t we buy them out? We’ve raised less than $1 billion, right? We can get to $10 billion while generating $4 billion to $5 billion of free cash flow a year. It’s a legitimate question.

I think if you’re post-C and strongly cash-flow positive, such that you don’t need to sell shares—in fact, you’re a net buyer, as you point out, either because you’re doing buybacks for employees or because you’re doing buybacks for founders—then you’re not trying to optimize valuations.

Going public might make sense, but I think you end up doing it for other reasons: mass liquidity, including your own.

But isn’t Larry Ellison kind of doing that, in a way? They just reported that he’s at 41% ownership of Oracle now.

No, that’s really funny. He’s buying out his shareholders every year with cash flow like we’ve never seen before, right? Why don’t we learn that lesson and do it before—not even bother to IPO?

That is true. Probably the beauty of it is—first of all, you’re right. Let’s talk about that now. For background, at Oracle’s IPO, I think Larry Ellison owned something like 23% of Oracle. Typically, that goes down over time. He now owns 41% of Oracle, right?

What’s he done? Every year, he’s run that business superbly. It’s got 43% operating margins, and he’s used that cash to buy back shares. He hasn’t sold any, so his ownership has just gone up over time. It’s exactly what you said. It’s a beautiful thing.

Really interestingly, 2 things happened this year. One is that the stock really popped 40%, and he got a lot of cloud cred. But the interesting thing is that this is the year he actually abandoned the buyback strategy.

Instead of taking all that cash and buying shares back, he’s taken all that money and put it into capex. Oracle was not free-cash-flow positive this year.

His own money, really?

Yes, he put his own money in, and he said, “No, we’re going to take this lovely, mature, cash-flow-positive software business and join the other crazy people in this capex-crazy hyperscaler land.”

The capex budget was, I’m winging it here, something like $30-something billion, and effectively they were free-cash-flow negative. The trick he used to get to this point is now not happening.

Luckily for him, it’s so clever. He bought when it was cheap for 10 or 15 years, then invested in AI, and got a 40% stock pop from that investment in AI just when he owned most of the company. It’s a thing of beauty, and it puts him firmly as the 2nd-richest man in the world, I think, for a period of time. You’ve got to love it. What a strategic mind.

Also, I think he looks phenomenal for his age.

Totally. I don’t know what he’s doing, but whoever his people are, I need all of them. I haven’t told you—I’m actually his blood boy. That’s why I didn’t sell the company. He pays me much more.

Yeah, there you go. I totally agree with you guys: $30 billion in 2024 into capex alone instead of buybacks.

What’s fascinating about it is that it’s such a different bet at a time when most people in their 80s are getting conservative. It’s like he took the Warren Buffett playbook—the Bible—for 15 years and did the cash buyback, like Buffett did at the Washington Post. Then last year he said, “Fuck it. I’m 80. I’m just going to double down here and switch strategies.”

It’s fascinating. If it works, it’ll be a legend.

Plus, once you buy a couple of Hawaiian islands, there really aren’t many things left to buy, right? You have to buy a planet.

Yeah. We talk about yachts, but Larry Ellison owns a big chunk of Hawaii. There’s not much left to buy.

Actually, a very famous billionaire once told me that once you conquer Earth, there’s only 1 place to go, and it’s space. That’s why we have Elon and Bezos. I always think of that. There’s always something, right?

Jason, you’ve mentioned Cluely. We have to talk about this company. Jason, why are you a fanboy?

I’m a fanboy because, if we look at AI for coding, we look at Replit going from $10 million to $100 million in 5.5 months, announced yesterday—that’s crazy. If we look at Lovable, it’s not far behind, and so on and so on. Cursor is another one.

What kind of bums me out is that, on the GTM side, on the sales side, I know everyone has made investments there, but they’re not as good. They’re slow to release features, they don’t work that well, and they’re just not as good as the developer tools for AI.

What I’m looking for is someone approaching this from a consumer level, with a consumer experience that could work for GTM. There’s no one. I love my old sales team—everyone I worked with is great—but overall, the sales reps I talk to for all the products I buy are terrible. They don’t know their product, they know nothing, and they add no value.

So they all need to cheat. All sales reps need to cheat because they don’t know anything, and they all need something like Cluely. I’m using Cluely right now.

There are a limited number of tools that do this in sales, but they’re either not real-time, they have an enterprise niche, or they’re glorified notetakers. This is what every sales team needs. Will it be Cluely? Maybe not.

But once in a while, like Slack or something, you need something to come up from the bottom to disrupt a market instead of coming from the enterprise. I just don’t see sales tools like that.

We can look at crazy things that are crazy successful, like Clay and others, but Clay is a 6-month deployment period with an agency that you pay $50,000 to. I want tools that you can use in 5 minutes, not—I’m not saying Cluely does all of it today, but I can already see hints of it.

If it could do 2 things today, we could use it. It would be the cheating tool for sales reps. It needs 2 features they could build in a month. That’s why I may be an intern. I was invited to be an intern this week.

It may take me a week to get up there, but you think I’m kidding. When I joke, there’s always seriousness in it, right, Jason?

You’re taking a way too academic approach to this, because the right tack—or the real question—is whether they’re going too far in their bid to get attention. Posting pictures with strippers on sofas, police cars arresting people outside of parties.

I thought that at first. When that initial stuff went out—when that stuff that we would call inappropriate went out—it seemed crazy.

If you look at that social-network thing he did along with Andre and the fundraising, that went like the other stuff went to other folks. Going back to The Social Network, it went to me. Okay, it went to me maturing the company just slightly, because there’s a whole generation of us in B2B.

Harry, this is where I’m a couple of clicks older than you. I can’t tell you what it was like when I took my team to see that movie. It was generational.

You don’t know what it was like when everyone was piling on Zuck, saying this was a terrible company, hoping it might be worth $1 billion. Today, what is it worth? We started this episode at $1.8 trillion, right?

His point when he did the interview was, “Listen, you guys on Twitter and LinkedIn are 2 or 3 years behind what’s going on on TikTok and Instagram. You’re in the middle, Harry.” He was like, “I’m going to bring some of that knowledge to Twitter and LinkedIn,” and I thought that was the social-network thing you did. I thought it was a 10.

You know, I want to come in on this because I actually read—I don’t even know who wrote it, because Notion’s got such a shitty UI. I can’t figure out who wrote the piece—but I got a piece sent around internally called “Leverage Beta Is All You Need: The LLM Business.”

It was so clever, and it gets back to this because there are 2 separate questions at stake here. One is how much should be steak and how much should be sizzle? How much should be core product versus marketing? That’s a general question. Then the second question is: are there certain forms of marketing that just go too far, which is Cluely, right?

You could argue the Harvey comment I made was that they did marketing. This was such a good piece. I’m going to read out a couple of lines from it:

“The brutal truth about the LLM business. Here’s what nobody wants to admit. When LLMs finally work at something, the implementation will be boring as fuck. Harvey isn’t some breakthrough in legal AI. It’s ChatGPT with a law costume. Lovable isn’t revolutionizing code. It’s Claude with pretty buttons.

“So, you have 2 choices. Option 1: wait until the LLM actually works, then scramble to build your ChatGPT wrapper along with everybody else who relies on the same thing. Option 2: start now while the tech is garbage. Lie about how good it is. Burn money on marketing. Claim the territory while everyone else is still laughing at you.

“This is leverage beta. The companies winning at leverage beta aren’t the ones building better products. They’re the ones who understood this dynamic first. The leaders lying about the present—11x or Icon—are arbitraging the obvious. Harvey or Lovable.”

It’s a great piece. What he’s basically saying is the models are getting better so fast that, even if you can’t do it now, you will be able to do it a year from now. So your choices are to wait for a year and compete, of course, along with everyone else, or lie, compete now, establish this kind of mental perception of the winner, and then collect the check when the time comes.

Going back to what I said, Harvey did that in a high-class way, because I don’t think you sell to law firms by hiring strippers. I don’t. Right? You could argue—you never know. Actually, I’m not going to speculate. I’m going to keep this thing highfalutin.

I think Cluely is clearly doing it in a different kind of way, and, as I say, we can discuss reputationally whether that kind of marketing works. But the meta-comment is: claim the ground with marketing and let the product follow on because it’s going to get there, because the models always get better.

It was a wise insight for a piece, and point. I would give him or her credit if I could just figure out in Notion where the author is. But there you go. Cluely might know. I’m going to ask—let’s see who wrote it. I’m going to ask: why didn’t you invest in this company?

I didn’t get a chance. I wouldn’t have invested in the company 3 weeks ago, or whenever Susa did, but I would invest in it right now that I get it, right? I’m not always that fast.

Actually, to be fair, my partner Paul picked this guy out when he was kicked out of Columbia and cold-DM’d him. I get it.

Jason, would you do this? Would you do 15 on 100? Would you do it? Cluely says the author is not named. Rory doesn’t know, and he doesn’t have enough data in Notion.

Right, show me that. If you can show your transcript to Cluely, though, truly, Cluely can look it up. Jason, would you do Cluely at 15 on 100?

I could do like 5. I don’t think I could get to the other 15. That’s too much risk. But I might do it. Yeah, I might do it.

You might do 5 from your fund?

Yeah. Yeah, I might do it if—listen, I’m making the problem. I get excited about companies in general, and then sometimes you meet the founders and it’s not what you thought, right? We’re all on our own journeys, and there’s probably a good chance what I see in Cluely is not what the team wants to build, right?

I mean, it started off as a cheating app, but I’ve just—I think all the B2B people are—I'm not, I don’t want to be mean. This is all I want. I’ll take the risk if someone can pull together an S-tier team in GTM. I’m in on it for real. They all claim they do, and they’re just pretty good.

We mentioned—I’ll tell you afterwards—we mentioned buying islands and being in Hawaii. The thing I just can’t get, and a lot of my companies are really perplexed by it, is the Slack lockdown, cutting access to it. Can you just help me understand: will it work? How do we think about what this actually means for Slack moving forward?

You know, all the leaders in B2B, I think most of them are going to circle the wagons and become more locked down. They have to be. They have to be.

When you’re sitting around the table, especially with a CRO and others who are under stress, what does the CRO want to do? Lock it down. Move to multiyear contracts and raise prices. That’s the strategy when things are stressful.

I think we can argue whether this is a mistake, but I think MCP is an existential threat within 12 months to every B2B company. Folks are going to lock that stuff down more because you can lock down your API, but when your MCP server is open, man, it’s rough. It’s rough.

I love what Zapier is doing. It’s running, like, 1,000 times faster now to become Zapier Prime because of this. They get it. HubSpot is figuring it out, but they were first, right? And Salesforce—I think most people do what Salesforce is doing, which is lock this down.

Yeah, I think that was helpful context, Jason, because my instinctive reaction is, no, you can’t do this. I’m a Salesforce customer. If you were telling me I can’t integrate in and out and access my data, I’d be miffed.

What you did nicely, Jason, is remind me of the duplicitous and sly ways that, especially the closer you are to having a monopoly, the easier it is to start locking stuff down. You’re exactly right: LinkedIn, obviously. Epic in the medical-record space is notoriously difficult to integrate with. You have to pay fees and all that.

It is interesting that, as you get big and defensive, you’re right, there is this instinct to lock it down. Part of me says they won’t be able to get away with it, that the customers will say, “Look, if you’re going to do that, then the value of Slack goes down so much to me that you can’t do that.”

My gut—and I think you said this to me, Jason, when I was talking to someone—is that probably this reverses to some kind of fee-based thing, where MCP access to my Slack information is a priced API call.

I don’t know if you can get away forever in a horizontal app like Salesforce with denying the customer access to their own content over the medium term. I don’t think it stands intuitively. I could be wrong, and you did well to remind me of other areas where they do it, but I just think it’s a sign of a decaying empire. It’s a sign that you can’t compete on the merits, and it’s a little bit of a dangerous sign.

It’s one of the signs that says this would be a good time for you, Mr. Customer, to consider your options, right? It’s like when PE moves in: price rises are coming.

That’s why I have the most respect for HubSpot, and especially Dharmesh, for being number 1 here. Launch-day MCP, OpenAI partner, ChatGPT partner, launch day, because I think it’s a threat to HubSpot. I think it’s an opportunity. Of course it’s an opportunity, right? Which is why they’re doing it.

But to embrace the threat—that’s badass. That’s the way you do it. You don’t—but, yeah, it is a sign of deteriorating everything, right? And as Slack deteriorates more and more, it becomes less and less our neural network. It’s going to get locked down even more, isn’t it?

Yeah. It’s hard to imagine a world where you say, “This is, as you said, the neural—this is the means by which we all communicate with each other, but no one can access that information for the use of AI.” That’s just not a thing. It’s not a sentence that survives.

So, final one: I have dinner with Benioff in London in a couple of weeks. What question should I ask him?

Taking Jason’s theme, this idea of agents, I would say kind of trying to get some data and dialogue around it.

How do you measure the efficacy of your Salesforce agents running on the Salesforce stack? How does that compare to third-party agents running on the Salesforce Slack stack? Are you better because you have the data, or are you worse because you're a little behind? Do you even objectively measure it? Do you understand, if you're using your sales agent, how you measure success?

We're all in—we have a wedge in the AI SDR space, and there's a bunch of others. How do you compare to them? If you're in Service Cloud, how does your agent compare to all the independent agents? Because to Jason's point, what would make it go faster is really simple. If the resolution rate on Service Cloud was 20%, you could only, let's be honest, eliminate 20% of your service center personnel. If the resolution rate with some third-party product like Decagon, Fin, or Sierra is 50% or 60%, then you're going to lose business pretty quickly. Are you measuring that? That's the question I'd ask him, and then I'd want to come to dinner. Do you want to come to dinner instead, mate?

I'll have dinner with Jason. That's a really interesting question. I'd be too scared. He might get mad at me and then cut off access to my Slack, and then I'd be screwed.

Right, we're going to do a Kalshi quickfire. As you know, this is an incredible betting platform that does predictive bets on world outcomes. We have: Will OpenAI accuse Microsoft of antitrust violations this year? Yes or no? The odds are 36%. You always have to state the odds; otherwise, it's just meaningless. Thirty-six percent means that if you bet $100 and it turns out to happen, you get $246 back. That's what it means, right?

So, I take that bet. Yes. “Accused,” by the way, is a wonderfully vague word.

And will they file a lawsuit and prevail?

Maybe not. But will Sam throw words out?

Yeah, that 2.5x on a yes, I'd take that chance.

No chance. I'll tell you, in my opinion, 0%. I'll tell you why: I think he already did. They already did, so I win.

Yeah, I think that's right. They already accused Microsoft of antitrust. They floated it internally.

Yeah, Sam. Everything Sam says that seems off the cuff, on the side, or a little futuristic, he's very clearly telling you what's going to happen. He's very direct. When you hear that they're thinking about it, he's done it. It's the same as filing. I'm not saying literally. I already think he's threatened it in a pleasant way.

The question is, does he have to go through on this threat, which has already been made? I think it will. I think it's enough to have said it. I don't think Microsoft wants to be sued for antitrust, so I think it's going to get worked out.

By the way, I completely agree. I think he's one of the most strategic communicators. The interview he did with Jack, who I love—Jack's great—but what brilliant timing for the message he wanted to land. He knows the message, which is Meta's poaching for $100 million. He's just put a dagger in the heart of Zuck's recruiting campaign to take from OpenAI. Brilliant.

He makes it feel like he's just sharing things with you, which he is, right? But I didn't get how clever his communication strategy is. I didn't get how clever it is, right? It's as good as anybody's—the best of anybody's, isn't it? Beautiful.

Okay, so the next one is: Will the US government take control of any AI company or project in 2025? The odds are $100 gets you $297 back if it's a yes, and $100 only gets you $125 back if it's a no.

So, I still think no. I think the push from the folks on the tech side has been very much AI for good, not AI to control it. I think all the very active tech people, from David Sacks to Andreessen Horowitz, have approached it as, “AI is wonderful and we should make lots of it right here in America,” not, “AI is dangerous.”

While it's pretty clear that the tech bros don't run the administration, it's pretty clear that the big guy runs the administration. My guess is this is just not important enough for the big guy to care about. So, thanks for the money, guys. On this, you can do what you want.

No, I don't think there's any impetus to say, “Let's seize control of Anthropic or something like that.”

No. So, even though I only get $25 more than I put in, I would take a no. I don't know, but watching what David says—which I think is very careful on the government side, very, very careful about what he says—kudos to him, right? I haven't even seen a hint of this from our AI czars.

If he knew it, he wouldn't say it, right? I mean, here's where having an ex-lawyer, very briefly a long time ago, in one of these roles instead of Elon probably helps. Sacks knows exactly what to say, but I feel like there'd be a hint if this were true, given that we're halfway through the year. But I don't know.

Agreed. It's far more likely to see some kind of regulation of Chinese AI companies. Not obviously taking control, but some kind of pushback there. I think that's highly likely, but not in the US.

Brief, short detour before the final one. Sacks had to divest a load of assets, including a load of crypto and also late-stage companies. Do you think he was hurt or helped by divesting? He divested in a pretty good period, a pretty buoyant part of the market.

I think it's hard to look at. The markets since then have been down but then back up. Overall—the facts—the overall S&P has been roughly flat, so no, not a gain or loss, but crypto's up. It probably cost him money, and again, credit to him. You don't have to like a ton about it to say he's doing public service. He's sold his assets to do that, and there probably has been a cost to it.

He's putting his money where his mouth is. There are famous occasions of people having to divest to join public service and then taking part in administrations that totally tank things. As a result, the divestment looks like genius. I don't think that's the case here. I think it cost him money, because look at crypto alone. Since the election, we've run up a little bit, then dipped down a lot for Liberation Day, and now we're back to roughly flat. It's been kind of a no-op.

Final one, boys. Trump Mobile smartphone: Will it be released before September? The odds are $100 gets you $716 back on a yes. $100 only gets you $108 on a no. The man moves at speed, boys. What do we think?

It's impossible. There's no supply-chain evidence of any phone in production. There haven't been any leaks of an actual phone in production, and there haven't been any leaks or signs of true product development other than a mockup of a golden phone.

I thought he was doing a product partnership with AT&T and was basically just sticking a Trump sticker on top of a different phone. If he wants it—well, listen, maybe I'm dated, right? The idea—the press I saw was that he was launching a phone, right?

Cluely quickly researched it for me while we're here in terms of all the supply-chain evidence. There's no notice of anything happening. If they're going to put a sticker on the phone, I'm sure he could do that yesterday.

That was a pretty good answer, though, wasn't it? It's pretty clear that Jason has—I mean, okay, Jason's actually making money. He's put money in, Harry. He's put money in. He's got $5 million in. He's talking up the stock.

But I will say, that was pretty impressive because it's sitting there. It's doing its recording.

No, no, I get you. I have to actually understand what's going on. Jason just has to be able to read. It's really lowering the bar for confidence here, people. I barely have to read. He just listens—clearly, just listens and watches the screen.

Yeah, and as long as you can read it back.

No, that was a win. That was a win.

As to the phone, look, I'm with Jason. There's no—if it ships, it's meaningless. It's not a thing. I'm not going to dunk on the attention span. There are a lot of initiatives in the administration that come and go. It's been a while since we've talked about Greenland, for example.

Look, this is a week when the administration had a big win. Let's just take it at that. I don't think the phone is going to be the biggest win.

I'll make a different comment. If I were Tim Cook and Apple, I would move heaven and earth to even make some kind of phone here, even a small one, just to do enough to say we're trying. I do wonder—I mean, that's a company whose business model is so exposed to China risk. If the solution on the making side is, “We don't make it in China; we make it in India,” that makes logical sense, but I just worry you could find yourself in the political crosshairs.

So, even trying to make some phones in America, even just 1, would, in my view, be a shrewd thing.

I agree with you. Make 1. Get Trump there. Get the picture. Get the marketing message: “Made in America.” Trump did it.

Please. That's exactly right. Make a little lot of 10,000. You'll sell them for twice the normal price, and you'll discover that Americans won't buy them, and you'll have tried. Make a good-faith effort. Get them off your back.

Boys, thank you so much for doing this. This episode was brought to you by Clo by Jason. Actually, Harry's house. It's Marley Boneley. Whatever. Actually, let's just do it at the 20 VC office cuz we can spill out onto the streets and then behind it.

That’s a great place to have a party, isn’t it? You’ve got 2 floors. What are the rules on spilling the party out into the streets? Is that cool?

It’s a private road, baby.

And guys, given the clear context, clearly party there tomorrow night. Ending with a comment that says, “Clearly, party, no rules” is not how I would suggest we end this movie here, guys.