Eventbrite以5亿美元出售，Databricks以1340亿美元估值融资50亿美元，以及SaaS为何像日本

Rory O’Driscoll

The quick answer would be, we’ll definitely talk about it, but it doesn’t matter. The interesting thing, Harry, is that that’s not the OpenAI story anymore. That’s the OpenAI story from a day ago, and the OpenAI story today is almost the exact opposite. It’s the code-red focus on the core.

It’s almost a statement that says, “All the other things we’ve been doing, we ain’t doing them now. We’re just going to be fixing our core product.” They’re even pushing on ads. They’re also pushing agents for healthcare back, pushing that stuff back and really going all-in on no distractions. So while I think this is an interesting announcement and we’ll definitely talk about it in a second, the zoom-out comment is that, 24 hours later, that’s not the zeitgeist at OpenAI anymore. The zeitgeist is, Google did a code red 3 years ago on them, and now they’re doing a code red back.

The one thing that’s kind of interesting about it—and this can be a little soul-crushing as a founder—is that there are only a couple of deals that matter to VCs. You’ll see a VC stay on a board for 20 years, and you’ll see them always hanging out with this one CEO. Thrive has a lot of winners, right? But this is a big winner, right?

Anything you can do to go deeper with those founders on your 1 or 2 winners, it’s power law on steroids. It’s parallel with what you do with your week. It’s power law with your deal flow.

This is OpenAI investing in Thrive Holdings.

Yeah, but it’s the same people, right? It’s still going deep on your winners. It’s, “I’m turning my VC fund into a holding company. I’m putting a billion or 2 over there,” right? And then I’m going even deeper with the number 1 company that I’ve ever invested in.

Rory O’Driscoll

First of all, I think it’s a great deal for Thrive because Jason’s right. The whole trick in venture is we try and pretend we matter, but in our hearts, we know our best companies matter, and the best marketing you can do is get as close as possible to your biggest deals and try to give them huge credit.

Put a bunch of money into OpenAI—in that, I think it was a $70 billion round—and we were there for Sam in the great fiasco of 2 years ago. My guess is he’s pretty darn loyal, and this is a chance for them to get a halo effect as they work on this initiative. We’ll talk about the specifics of the initiative in a second, but it’s a huge halo effect because, as your last speaker said, you go in with the OpenAI moniker and you get some real attention. So it’s awesome for Thrive.

1. Databricks Raising $5BN at $134BN Valuation: Cheap or Not?

The advantage to OpenAI is much less clear. I’m not sure they’re putting in money. In one of the press releases, they said they’re going to get a lot of specific data from some of these verticals that Thrive is pushing into, but my guess is I know who was ecstatic when that was announced and who was like, “Yeah, whatever.” That’s pretty clear.

We said something about the importance of our winners, and Rory, you humbly said that it’s all about our star founders. Databricks is one of the stars of the last generation—or this generation. They’re rumored to be raising $5 billion at a $134 billion valuation. It’s 32x 2025 sales, which are $4.1 billion. They’re at 55% year-on-year growth. Is this actually cheap? How did we analyze this one?

Rory O’Driscoll

Well, I wouldn’t call it cheap, but reasonably—I mean, look, possibly reasonably priced. It’s very convenient right now because the direct competitor, Snowflake, is public at roughly the same revenue, around $4 billion, growing at 28%, and valued at $80 billion. So, 20x revenues, right?

It poses very nicely the big-picture venture question: how much extra in multiple do you pay for how much extra in growth? Right here is the worked example. You can buy a profitable company doing $4 billion with 28% growth at 20x, or you can buy an unprofitable but faster-accelerating company at 32x or 33x. Databricks is growing allegedly at 55%, not 30%, not 25%, so is that extra 25% to 30% of growth worth it?

Do you agree that’s fundamentally the question you’re asking: how much extra revenue multiple do you pay for how much extra growth? Would you agree, Harry?

I agree with that.

Rory O’Driscoll

I think what you very quickly do is the math. What you say to yourself is, if that extra growth lasts for any length of time, extra growth’s worth a hell of a lot, to use a technical term, because that compounding keeps going. So if that growth persists for 3 or 4 years, then maybe that extra premium is worth every dollar and then some.

Then you start saying, how much extra for how much extra growth? You kind of look at the public markets to figure it out. Then you discover something really funny, and Jason’s talked about this: there is literally only 1 public company growing more than 30%, and that’s Palantir. For the record, that’s growing at 50%, wildly profitable, and valued at 80x sales. So you just don’t have a data set publicly to assess this.

I think the simple answer is it would be the second-best public company if it were public today. It seems about right. The crazy thing is it continues to modestly accelerate.

Rory O’Driscoll

Yes.

It’s just something that we haven’t seen before. It’s something we all have to adjust to: that you can continue to accelerate at this scale.

Rory O’Driscoll

Well, it justifies all the craziness we see in venture, at least for now, because the headroom is still there.

I totally agree, Jason, and I’d forgotten that the reacceleration changes everything. I was going to say, when you develop that model of how much extra 55% is worth versus 25%, you make some assumption of gradual deceleration and gradual conversions, because that’s the only rational thing to do. Then you can come up with a number. It’s a high revenue multiple, but still a number.

You’re right, Jason. When stuff starts reaccelerating at scale, it’s almost hard to figure out the model. By definition, if it continues to reaccelerate, it’s infinitely valuable, because that’s just what the math says. It’s probably not going to be infinitely valuable, but it points out the power of reacceleration at scale. If you could even stipulate going from 50% to 55% to 60% at scale, oh my God, there’s huge value.

It’s actually the same dynamic. That’s why it’s hard to value the big foundation models. When Anthropic went through that bout of reacceleration this year at scale, everyone realized the model was wrong. They had to raise their estimates and raise the value, and that’s why you saw that step-function increase in valuation, because reacceleration is really hard.

We see it maybe in 1 in 3 companies for a single year. Only 1 in 10 does it for 2 years. We very rarely see reacceleration when you’re already at 50%. For God’s sake. No, you’re exactly right, Jason. That’s the killer fact here. That makes it hard.

So you end up saying to yourself some version of what you said earlier: how big is the TAM? Because in the end, the only thing that stops something that’s reaccelerating at scale is when you hit the wall of, “Well, you sold to everyone.” It’s the Zoom thing.

Yeah. Just also remind you that seeds are for suckers.

Rory O’Driscoll

When you actually look at the certainty that you have that Databricks has a 3–5x from here, you’re absolutely right. When you think about the opportunity cost of, “I can put my money here or here,” risk-adjusted and time-adjusted, you could make a very coherent case that it is a better deal to put your money into Databricks, or like a client of Perkins did with Anthropic at $180 billion, than it is to put your money into a much less certain Series B with a 7- to 10-year duration from there.

Yeah, or just a seed VC deal at a $60 million post on a SAFE. It’s just hard. The Databricks deal seems like a better deal.

Rory O’Driscoll

$60 million post? Yeah, it’s quite cheap for a seed, honestly. It’s getting worse.

Snowflake’s up year-to-date 60%, and we’re seeing the reacceleration of Databricks. Do they just both grow into absolute monsters? Can they peacefully coexist? Do you think one takes majority market share? I had Ron Gabrisko, their CRO, on our 20Sales podcast, and he said Databricks’ technology is 5 years ahead, which I thought was a really interesting statement.

Do you think they peacefully coexist? Does one take a monopoly? How does that look?

I mean, Ron was correct.

They came from slightly different places at slightly different times. Snowflake originally was very much your SQL data warehouse in the cloud and was a few years earlier than Databricks, which was originally, I think, the Spark product and all about moving data and more AI use cases, even out of the gate. So he is correct in that. No one's going to coexist peacefully.

They probably hate each other. In fact, we know they hate each other because they time their sales events to overlap with each other. I would say, at different times, they're not going to quote-unquote peacefully coexist. They're going to struggle and fight against each other for the next 10 years, just like SAP and Oracle fought against each other for the last 20.

They're going to want each other's lunch, eat each other's lunch, and it's going to be a grind. I don't think either of them folds from here. It's hard to imagine the core value of a relational database going away for transactions. So, the Snowflake asset is money good, and separately, I think Ron is correct: Databricks has more of an advantage in brand-new, AI-centric data manipulation and data movement applications.

They're in bordering, adjacent, overlapping markets. They want a bit of each other's market; they're just going to slug it out. It probably means that at some point some of those margins get dinged a little bit, but you've seen it before. You've seen it, as I say, with Oracle, Sybase, and Informix. You've seen it with Workday, SAP, and Oracle.

This is just what I mean: Jason knows most enterprise software markets tend to be oligopolies, and they tend to punch each other for 10 years.

In our little corner of the world that is scale-ups and startups vis-à-vis Snowflake, the one thing I do know is we're just starting to learn what we can do with our data with agents. We're just starting to learn, and even Snowflake is still learning, right? It's new to Snowflake.

When we had the CEO of Snowflake at SaaStr Annual in May, they were just starting to talk about how they were going to use agents, right? Now, fast-forward to today, a couple of months later, all the vibe platforms can directly access Snowflake data.

Yeah. Literally, we could fire up Cursor, Lovable, or Replit and just build an app right now while we're here and access our Salesforce data. What will that mean for how we access that data? What will that mean for how we think about CRM, how we think about where we host, and how we access it?

I'm just not smart enough to predict what that means in a year, because that wasn't even possible a couple of weeks ago. Can you access every bit of data in Snowflake or Databricks in a Fortune 500 company? I assume the answer is no. I assume a lot of that data is restricted, but that is an epic change that I can't even predict the slope of the curve for next year.

What happens when I can use easy-to-use agents, where I don't need a lot of engineers, to access all of my data any way I want—to build any report, any link, or any workflow from that?

I think we're 1% on this journey.

The amount that will empower our data with agents—we're just learning.

If I were to push you honestly and ask the question: Will we see your CRMs, your Salesforces, your HubSpots—will they become databases which agents sit on top of and feed off? Or will these platforms move with the agentic world, build their own, and not become sheer databases alone? If I push you, which one is the likely outcome?

I think the fact that Marc Benioff has put 2,000 people on Agentforce tells you the future right there. He's already got 2,000 people on it.

Well, it tells you his intentions. It doesn't tell you his ability.

We are still early, enterprise-wise. I think if you look at Salesforce in particular, they've got 2 years to unlock all this capability. With the rate of change, Marc has to do what he has to do, because 2 years is not a lot of time at Salesforce.

Traditionally, that's like a major release, right? Snowflake turned out to have time. We might have thought Snowflake was struggling at the start of the year, right? We don't think Snowflake is struggling anymore, so there is more time in the older enterprises.

Where that breaks for those that have massive amounts of data, I am not smart enough to predict. I would not bet against anyone managing a huge amount of structured and unstructured data. I wouldn't bet against anyone.

What will probably happen is, if you want to build agents that are just using your CRM data and you're a Salesforce shop, just like Microsoft was able to bundle Microsoft products, you'll probably take the agent from Salesforce and get a combined thing.

But I think, and to Jason's point, enterprises that want to build more powerful agents—where they're not just drawing the data from one app, i.e. Salesforce, but drawing the data from maybe 5 or 6 different sources—the more likely architecture for that is some version of stuffing it all in Snowflake and then running an agent directly against that.

As you were talking, Jason, earlier about what you're seeing with Snowflake, I remember one of the things we did 5 or 6 years ago that turned out in retrospect to be very smart was literally opting to stuff all the data in Snowflake across all the systems. Over time, you have access to that.

So, if I'm a large enterprise—let's say I'm a bank—and I want to use my deposit system plus my Salesforce system plus something else, at a certain size it won't work easily in just Salesforce. You'll want to go, just like every large enterprise since the dawn of time: Some apps you'll want to build yourself.

If agents become strategically important enough for super-big companies, then they'll throw $5 million and a bunch of Snowflake and Databricks at it and just build it themselves. So, there's no doubt that agents can get a fair slug of the market, but I'm also sure that at the high end there'll be a bunch of wonderfully bespoke projects that will make systems integrators rich for the next decade.

Everything we're doing right now in AI is very exciting, but I've been thinking a lot about how Gainsight's been locked out of Salesforce for 2 weeks. I've been thinking a lot about how OpenAI just kicked Nick's panel [?] permanently off OpenAI this week for a security breach.

I really think that with agents running everywhere with our data, the folks that can securely manage that data and the folks that have secure, or seemingly secure, agents may win. We may be willing to bend some rules in the age of AI. We're willing to bend some rules to move quickly.

I'm not sure. I think security is going to benefit the incumbents. I think we're going to be worried that our agents are depositing our data in 100 different places, and we're going to be worried. It's crazy. I love Nick and the Gainsight team, but essentially their app has been down for 2 weeks with no known resolution time, because Salesforce has kicked them off their platform.

They kicked Drift off 5 months ago. Drift will never come back. It is dead. It is completely dead. I think about those things, and I think about where all these agents are taking our data. I might want my agents from Salesforce, Snowflake, and Databricks.

When things change, we get very excited about new vendors, right? Because the incumbents can't do it, we get very excited, and inherently we take a little bit of risk. We try to contain that risk in a pilot or in a less critical source of data, right? We always try to measure the risk at the beginning.

I wonder whether this is our sensitive data flowing. It may not matter for your average startup, but I generally am worried we're underestimating security and data residency in general. I just worry, and I feel bad, but I couldn't imagine going through this at Gainsight.

Do you think the world, and big companies, are more unforgiving now? Over the last 10 or 15 years, a lot of companies have had breaches. Do you think this was more serious?

It is more serious.

Yeah. They literally—so, Drift had the security breach. 700 folks' data were downloaded, and now a pirate group is asking for millions of dollars for each instance. Then this happens again with Gainsight.

I can't speak for Salesforce or the team. I would be more conservative with who I let touch my data. You could say it's an OAuth issue, whatever, but I don't want this ever happening again.

One time, we can blame Drift because it got acquired by another PE firm that got acquired by another PE firm. But 2 times, I might start locking down my platform. The third time, I might say I'm just going to own all the agents.

I'm done with these risks. I don't know. It's getting worse when people want to steal your data, sell it, and ransom you for $1 million across 700 orgs, including Cloudflare.

How big are the SecOps teams on most startups you work with? 50? Oh, thank you very much. 100. How big's the latest deal you did? How big's the security team?

The funny thing is, Jason, both the examples you cited are mature, first- and second-generation SaaS companies, not brand-new, AI-first companies. What you're saying is new companies might get tagged with the consequence of a more restrictive security policy, even though you could argue it was older companies—companies that, I think, in both cases were PE-managed—that actually caused the problem.

Life is unfair, but there you go.

Probably just before we started doing this, it kind of felt like the enterprise was going to win, that ServiceNow and everybody was going to win—and then it doesn't feel like that at all. Then we saw the numbers take off, and basically very few incumbents grew materially this year, right?

But this could be a piece of their revenge of the enterprise.

Yeah. Are you cynically suggesting that a large enterprise software company could say, “I’m using security as an excuse to cut you all off, but lo and behold, I have my own agent product right here, which you can now safely buy, Mr. Customer?”

Rory O’Driscoll

I think it’s a good excuse. If you don’t want Glean slurping all your data, right—which, if I’m Glean, I get it. If I don’t want all of it going out of Slack or whatever, many incumbents, I think, have the best excuse there is because the existential risk is just too much to the vendor. It’s not worth it.

I don’t know how many board meetings you guys have been in where the first half of the presentation was about how we’re going to be more secure in the age of AI, but for me, it’s been close to zero. It’s been close to zero in the above-the-fold part of the conversation.

2. Eventbrite Acquired by Bending Spoons for $500M

So, we spoke about data breaches’ growth. I do want to take the flip side of that, which is a less positive or optimistic side: PagerDuty, 2x, a $1 billion valuation, at $500 million ARR and 4% growth. We just saw it—literally, it’s not on the schedule.

This is where Rory gets nuts at me because I just add shit without asking him and then expect him, on a whim, to come up with something. Normally, we can edit out pauses, but now he’s going to be extra pissed with me. Welcome to my life. Eventbrite—they’ve been acquired for around $500 million.

Yeah, that’s a premium, right? So, that’s 1.5 times revenue.

1.5 times revenue.

With a 50% premium.

With a 50% premium. So, new news in today. We can take PagerDuty because it’s so fresh, having that Eventbrite news, that it’s hard to formulate thoughts. But how do we think about this bluntly, very harsh new reality, given the lack of growth and the subsequent pricing from it?

Rory O’Driscoll

I’ll surprise you, Harry. I’ll cover both, despite the complete lack of notice. I think there are 2 separate things, and one perhaps is a positive, right? One is kind of a fact about being public. When you’re public and your stock is floating around at a lower valuation, you’re just very vulnerable to this. Someone comes to you, you’re not growing quickly, they offer you a 50% premium, and you get called in. The board gets called in, the lawyer gives you the speech about fiduciary duties, and if you can’t come up with a convincing reason why you can build better value than that premium, you’re forced to take it. It’s not 100%, but it’s a tough place to be, right?

I can imagine the conversation at Eventbrite. I can imagine the conversation 2 weeks ago at Semrush, and I’m sure PagerDuty are thinking about the same thing, right? But now there are 2 things to say at a wider level. The first is, let’s start with the positive: someone else—very smart money—thinks these things are worth buying. They’re looking at it and saying, “You, Mr. Seller, haven’t created value here, haven’t found growth, and I think we can.”

3. Pagerduty's $1BN Market Cap, Just 2x Revenue

If you look at the 3 companies we’re talking about—let’s lump in Semrush, because they were acquired 2 weeks ago by Adobe—Jason and I disagreed, but we definitely felt someone like Adobe could do something with that asset. I’m not sure what direction Eventbrite could take, but PagerDuty, you said in your notes, Harry, has an obvious set of next products, including AI agentic products around downtime resolution, that feel obvious to me.

Frankly, I wouldn’t be surprised if an aggressive PE firm said, “Oh my gosh, I can buy this thing, I can then buy some small, hot AI startup, put them together, get this thing back to 20% growth, value it at 10 times, and look like a hero.” So, the positive would be that other people—smart, savvy money—look at these assets and say, “2 times revenue is stupidly cheap. I’ll have that.”

I hope so. I hope there are more deals. We see General Catalyst and others doing this: let’s add AI to services businesses. We see a lot of talk of this, but right now we’re not seeing a lot of these mashed-together legacy companies at hundreds of millions in B2B and hot AI startups magically flipping it into a 20x, 15x ARR company.

I’m not saying it’s not coming, but we haven’t seen PagerDuty.com and PagerDuty.ai magically mashed together into a winner yet, have we?

Rory O’Driscoll

No, you haven’t. And you’re right, Jason. But the funny thing is, I like PagerDuty. We looked at the deal 10 years ago. My then-new partner wrote a term sheet. We should have let him pay a little more because he was right. We love that market.

I remember the investment memo from 10 years ago, and it said clearly, in summary, “This is a great market. It’s going to tap out, and you have to add a whole series of add-ons around managing the process of downtime or security breaches, managing the process of getting something back up.”

We didn’t have it at the time, but now, obviously, adding AI-enabled operational resolution, right? The direction was clear and is clear. So, you’re right, and they haven’t done it, but I think that’s just a disappointing outcome, let’s just say, right?

I think the direction of travel should have been clear, and if you’d been able to add it, I think you would have. We talk a lot about how distribution is a huge advantage in software. Literally every ops team on the planet uses PagerDuty. For God’s sakes, it’s pretty obvious what to add here. People, get it done.

Rory, have your views—and Jason, too—changed on market size as an investor? Given what you just said, every ops team on the planet uses PagerDuty, and it’s a billion-dollar valuation. A 10% holding would be $100 million. It’s just a very sobering reality.

Has your view on market size changed when investing today, given where else you can put your money and the sizes of those markets that we’re seeing with your Lovables and your Replits?

I see more and more folks who I thought would grow out of a small TAM not grow out enough. When I started as a B2B founder quite a while ago, everyone seemed to grow out of small TAMs, for the most part—some better than others—but it felt like we had time. It felt like you had 4 or 5 years to figure it out at each stage, and you could see it coming.

Now it feels like, now that the average public SaaS company is growing at 16%, no one figured this out. We literally just looked at this chart in my presentation before we looked at this. No one’s ever grown this slowly.

4. The TAM Trap: Why SaaS Is Like Japan

So, if the public companies can’t figure this out, I know the next generation of kids should do better than the adults, but I’m worried that the majority of public SaaS companies didn’t figure this out. The majority of public SaaS companies, I think, are in a TAM trap.

Rory O’Driscoll

It’s a good title for the book.

The TAM trap. TAM trap.

Rory O’Driscoll

The TAM trap.

We love them, right? How did the Aaron Levies and the Drew Houstons and the others not figure this out? And I love them, right? How did we not all figure out the TAM trap? What hope is there for the rest of us?

Rory O’Driscoll

I can give you a clear answer on that. First of all, I think it’s a great division of the discussion into the TAM trap for the existing companies, what it means for the new AI companies, and then, maybe third, what it means for venture.

Let’s start with the first one: the SaaS TAM trap. You keep saying, Jason, “How did they not figure it out?” Let me just offer a different perspective. Maybe there’s no answer. In other words, there are so many SaaS companies. It’s not that everyone was an idiot and couldn’t find the market. I don’t believe that at all.

I believe that we made so many companies that we saturated the markets, and by the time you got to the point where you needed to expand beyond your market, in many cases there were other venture-backed SaaS companies in the adjacent market, so you just kind of ran out of room, right? In fact, I did a blog post on this in 2019 called “Hunger Games and SaaS.”

That’s my opinion on that. For many of these markets, it’s not that some of the CEOs you cite are idiots; it’s that you have high penetration of the markets. There’s not a lot of room. I mean, you take the quintessential one you and I talk about all the time, Jason: Zoom. Everyone who needed a Zoom account has one, and everyone who has a Zoom account has a Teams account, the poor bastards, and they’re done, right? There’s nothing more to sell. You’ve got to build a new thing.

The initial, quote-unquote, obvious new thing for Zoom pre-AI was that whole contact-center business, and they couldn’t get that acquisition done. There were already incumbents in the space. So, I think pre-AI, you ran out of time, right?

Harry, to your question, as we’ve said many times, overpayment only works when the TAM is huge.

True.

Rory O’Driscoll

In finite TAMs, you’ve got to bid more tightly, right? I think it’s so funny. I’m going to give an anecdote about PagerDuty. When they went public and we did our internal autopsy, because you do, right, I looked at the model we’d underwritten 5 or 6 years ago, and we were accurate within 3% on the model prediction on revenues. All that happened was the market was just willing to pay more for the asset, and now it’s not, right?

So, on the existing space, it’s the time constraint, and I think you just have to be careful on price. The question on the new markets, which I think is believable and credible, is: Is there AI—we’ve discussed this a million times—does the AI labor expansion save us all and allow us to reach higher in price and still get these huge TAMs, or are we going to be in the same place 8 years from now? That’s actually the multi-trillion-dollar question, as it turns out.

Well, I think there are 2. One is: can AI allow us to tap more into labor budgets, right? I think the second one, and I was trying to summarize this earlier today, is: can AI provide so much value that you can charge an order of magnitude more than you could charge before?

Gamma charging $100 a month instead of the $8 I pay for Canva, right? Or Cursor charging $500 a month when I pay $3 for Jira.

And so you wonder, can we do that? And, you know, we don't talk about Zoom much. When I think about Zoom, I can't think of a better technical founder running a leader like Zoom. I can't figure out someone I respect more on every level—as a human, as an engineer, as a leader—than Eric.

Why, in 4 years? This is a mean question, and I don't deserve to even ask it. Why didn't they capture more TAM? Why didn't they find a way to add 4 billion of notetakers when there's a trillion notetakers? Why didn't they? And I don't know the answer. It's not because you don't have one of the smartest people in the industry thinking about this for a decade, but there was no great second act yet. There's no great second act. Going back to this, I worry, and this is why I tell founders to take their exits and then say no because they go bigger—but by default, take it.

So many things in that to unpack. One is the multiproduct thing. I think one of the big takeaways we've had is the need to be thinking about that second product much earlier than you would have thought. You don't want to wait till you hit the TAM trap. What was the word or expression? Yes, I loved it.

It was good, but I'm forgetting. What did we call it?

Was it trap?

The TAM trap. Don't wait till you hit the TAM trap. We were thinking, actually, we're comparing 2 of our portfolio companies, and it's invidious to name them.

But we would say one of them has compounded really well because it's continually added a new product that for the first year or 2 is a couple of million dollars, but layered it in, and now it's many hundreds of millions. So, I think watching that TAM trap is key.

But the other thing he said, and it was in the speaker notes too, is, you know, on the AI pricing, if your pricing is versus labor or value created, you're getting enormously great prices because you're saving a lot of labor. And I think he said in the speaker notes, “What happens when there's 2 or 3 of these companies and the competition goes from, ‘Hey, I'm saving you $1,000 of labor a month,’ to, ‘Yeah, I'm saving $1,000 of labor, but there are 3 providers of the same AI software, and they're all willing to do it for $100?’”

So, your ability to get $500 gets eroded. Do you start seeing that happen in AI very quickly? I mean, it hasn't yet, I think.

Speaking of the pricing challenges that we have here, we have Workday coming out saying seat reductions are an existential threat. We had Jeff Lawson from Twilio on the show with the 3 of us, and he said that we are unwaveringly going to see the movement away from seats, and that is going to happen. Can companies still price by seat in the age of AI? Is Workday inherently threatened and right to be concerned by this existential threat?

You know, when Jeff—if folks haven't watched it, it's worth rewatching—he was so good, right? And he made that comment that Twilio, especially if he were still CEO, would have been somewhat insulated from that because it's based on usage, right? And, in fact, Twilio has seen a little bit of a resurgence, right? It has seen some reacceleration.

And he said he was very worried about seats, but he said, “I hadn't been in the game in a little while, and I've been doing AI and working in my shop.” So, I didn't fully get it, but it has resonated in my mind since: he is right, and everyone is shrinking headcount in tech, at least. Even if they're not shrinking headcount, ARR per employee is going to keep going up.

I crunched all the data before this morning. Everyone is going up: HubSpot, 2.8 times more efficient than 2021; Salesforce, 2 times. Microsoft has said they're already past peak employee—permanently past peak employee. We're all going to figure out how to get more ARR per employee. And if you're a leader, you're just going to run out of seats, right? Mark was kind of aware of this when he did the pod, too.

So, I don't have the answers. In the early days, it probably doesn't matter, right? Your model is your model, and seats work well in some places, but it is existential. We are just going to get more and more efficient. One of my biggest worries for investments is when startups aren't getting more efficient. I'm not talking about profitability. That's an investment. I'm talking about where their teams get more bloated as they scale.

I kind of am out. When I go to a board meeting and a CMO says, “Well, I could do that, but I need 50 people,” or a product guy says, “The reason we're late is I need another 80 people on the product and engineering team,” I think it's time to part ways. Give them a nice package and a good recommendation.

Can we get AI to replace you? I want to see you grow 100% next year with 50% headcount growth.

Yeah. I think that's healthy today. The way through 2020, early 2023, was, “I need 200% headcount to grow 100%.” And there's still a lot of that DNA in the ecosystem. It's still ricocheting around, probably in the majority of executives people will talk to.

I don't think the seat is dead, but as time goes by, I get more and more worried that it feels like Japan: our population is organically shrinking. SaaS has become like Japan.

It's a great economy, but if everyone only has 0.9 kids, I mean, there's only so many seats to go around.

I want to come back to the last comment first, almost—your comment on 2-to-1—and you're exactly right as I think about it. I hadn't thought it until you said it, but in 2021, every discussion was some version of, “Look, to get that extra 10 points of growth, we're going to be twice as inefficient at the margin as we were overall.” And the result of that is overall efficiency has deteriorated.

What you're saying now is, at the margin, we're trying to be twice as efficient, not inefficient. So, you're exactly right. Everyone's efficiency is creeping up. I think that's obviously, independently, totally correct and super insightful when you said it. It might explain why, even though the growth rates of the public companies have gone down significantly, the valuations and the revenue multiples have not gone down as much.

And I think it's because, in return for slower growth, SaaS companies are at least getting wildly more efficient. It's not 1-for-1, as we've discussed, but at least it's better than nothing, right?

But then, going back to the existential threat from Workday, I think software prices are based on value delivered, right? And when you couldn't measure value, all you had was per-seat pricing, and therefore that's what people went with: everyone has to access the software; everyone pays so much, right?

Then you had usage pricing, starting with AWS, which was inherently a more rational, for buyer and seller, way to allocate value and track value more closely. And I think the trend in everything is just to get more efficient. I think that's how capitalism works.

So, to your point, if you're using AI to deliver value, and if the AI is doing the work, it's going to be super hard to have a $3-per-seat model because it's irrelevant. Now, I don't think for a lot of Workday it will be as irrelevant as for some. There are other areas, like, for example, some areas of Salesforce. You can imagine, to the extent you entirely automate an SDR team, it's going to be hard to have a per-seat SDR model.

My guess is Workday will still be able to have some kind of X dollars per month for end employees served and then Y dollars per month for a platform fee for HR staff actually using the software. But there probably will be fewer HR staff using the software, to your point, Jason, because if half the make-work that HR was doing is now done by AI, you're not going to get the same price.

Now, maybe you have—so, therefore, maybe you have to charge on the value delivered versus simply the seats, and that's a more complex calculation. And I think that is true: all our companies are wrestling with this. You deliver a ton of value in AI, and maybe now you get the innovation budget; no one cares. But a year, 2 years from now, you're going to have to link, as a startup, your pricing to the value delivered.

And you're going to have to measure. And it's a lot harder to measure value than seats. You can count butts in seats pretty easily. Every login is a butt. When you're trying to measure value delivered, that's tricky.

5. Lessons from Companies Hitting $100M ARR

Can we just go back to the growth and efficiency element? I always think when you're creating content, you have to think of your customer. And the customer that I always have in my head is the founder on their way to work or the operator on their way to work listening to our podcast or watching it.

They're hearing, “I want growth, growth, growth.” And now they're hearing, “I also want efficiency, efficiency, efficiency,” and $2 million per employee for Gamma and Lovable, at whatever it is. Do we just need both now? And is it a higher expectation to meet the bar for VCs?

Or, when you are looking at those 2, is it, “I want growth above everything, and I'm fine to see less efficiency in the early days”?

6. The Relevance Game in Venture Capital

I think in the fastest-growing companies that I've invested in, no one gives a rat's ass about the bottom line. That's a different metric than how you scale today, right? And Rory made this point. We all made this point before, and I made it earlier: there are actually fewer and fewer companies that are true outperformers.

So, as soon as your top portfolio company outperforms, everyone wants to give it $100 million today at every board meeting. And no one really cares about efficiency per se as long as they can get their money into the deal.

So I don't think that's the issue. I think we're confusing the fact that everybody is just generating more revenue per employee. But you're going to see this—we've talked about it before—with this ICONIQ data earlier in the year: the fastest-growing AI companies, even with high inference costs and R&D costs, have the lowest burn multiples because their revenue is growing so much faster than their inference costs. I think what we're hoping for is that companies have never gotten to $100 million ARR more quickly. If you do, I don't think we care how you get there anymore. I don't think that we care.

But we know deep down that hiring 1,000 people isn't the way to get to $100 million in a year. You can't hire them that quickly. As Maggie said, they're not all going to be great, right? So you literally can't brute-force $100 million in 10 months with humans. Maybe Larry Ellison or Marc Benioff could, but I don't think anyone else could. There's just not enough calls, and you can't go from $1 million to $100 million in 10 months without massive inbound demand and a lot of AI.

I think you should break it up into a couple of areas. And in fact, as we think about it, Jason, let me get your insight on your eternal question on employment, because, big picture, it's divided up into AI startups and mature companies.

Harry, I think the comment on ARR efficiency per employee is very much a mature-company comment. If you're a public company and you're only growing 10% or 15%, you better be kicking off cash or you'll be in trouble really quickly. Even if you are kicking off cash, you'll still get grief. But those are the companies that are optimizing their ARR per employee and are just focused on FCF, free cash flow, right? And that's a very different set of people with a very different set of dynamics than the AI startups, and we'll come to them in a second.

So, public companies are grinding on efficiency, which means, as you say, lower employment and all that stuff. Now come to the private AI companies. I think there's 2 categories. There's a small number of companies that are taking huge amounts of capital because they need it for model development, and primarily it's not humans—they've got to spend it with NVIDIA. Look, no one's telling OpenAI to be efficient, or if they are, they're clearly not listening. To a rounding error, those companies are, as Jason said, able to get all the money they want, spend it on compute, have a relatively small headcount relative to their size, and no one's saying, "Be efficient." They're just saying, "Grow quickly."

Then separately, at the apps layer, you're seeing something slightly different, and Gamma is a good example of that. The interesting thing is, because of this amazing new capability, for lack of a better word, called foundation models, there are people in apps land building a product, shipping it, and getting such traction that the traction is ahead of their ability to hire. I mean, literally, there's no way to spend the money.

7. The Future of Labour Markets is F

We're seeing some of these app companies be astonishingly capital efficient, especially at the early stages. Gamma is a great example of that. It's like, you ship the product, it's freaking amazing, people buy it, and they give you credit cards. By the time you get around to hiring a sales force, you're doing so much revenue already that you're kicking off cash. At the apps level—not all the time—I think some companies at scale are spending, but we're seeing the combination of hypergrowth and reasonable margins. They're not as good as SaaS, but still 50%. If you have that, then you have quite an attractive profile, right? Not all of them are the case. Obviously, the coding companies have margin issues, but a lot of the companies are getting a long way with not a lot of capital and definitely not a lot of employees.

So I think those are the 3 categories. And the interesting thing, Jason, this goes back to something you've been talking about, and I've been trying to figure out the answer to what's going on with employment, right? What's the consequence? I've been more, "It'll all be fine in the end," and I still stand by that. But the interesting thing, when I listed those 3 categories, is that the 1 thing they all have in common is they all don't need people. The big companies can't have people because they've got to be efficient. The model companies don't need people because they just need geniuses and GPUs. And the small AI app startups are going so damn quickly they can't hire people. That's not great if you're people.

It's not.

It's not. And overall, I'm an AI optimist. I think all this unemployment thing is bullshit. But in the near term, what you recognize is why it's a toughish market for tech startups—for employees in the tech marketplace—because, in the labor-versus-capital discussion, you need more capital relative to labor at the moment.

The topic we have to discuss: we talk about Replit the whole time, but then Google has come out with a competitor, and we've always been waiting for Google and ChatGPT to come out with one. It's been very good. It's tied to Gemini, which has obviously blown past a lot of people's expectations.

When we look at this, how do we analyze this? Is this a case of an incumbent waiting for enough traction in a market and then going, "Thank you very much, Mr. Startup. I'm going to come in now with great models and distribution, and it's game over"? Or have they actually left it too late, and Lovable and Replit have built enough user base, enough brand, and enough brand trust that there is still a real dominant threat to Google's new product?

I did try it, for what it's worth. They launched this week. They launched a Replit-Lovable clone with no database and no OAuth, so it's really—it is what it is. They said it's coming soon, and sometimes that's okay for big companies, so we'll see.

One thing that hasn't changed in the age of AI is that big companies only have so many priorities. They can introduce a lot of little tests, but at the end of the day, it takes a lot of energy in a big company to keep a big initiative going because there's so much else to support. So we'll see. It wasn't impressive in itself. But, on the other hand, it only took Google less than 10 months to launch their competitor. You don't get 5 years anymore. I mean, Datadog just launched their PagerDuty competitor in the last 24 months. When was PagerDuty founded—2008? You don't get that much time now. Now you don't even get a year. The incessant pace of cloning and competition does worry me, man. If you only get months before the big guys come into your space if you blow up, it should make sense, right? If you go from $0 to $200 million in a year, you should attract some competition. But it's not a free lunch, right?

Totally. But is there something that we take from this? Rory just did Glean AI, I think it is. We're like, okay, they're not going to go into Glean AI, but they are going to go into Lovable, they're going to go into Decagon, they're going to go here. There are themes where they're like, they are going to go, and there are themes where we're like, dang, model providers—the core question being, hey, where will model providers go in the application layer and threaten our businesses? I mean, competing with coding tools is not that big of a jump, with whatever happens.

I like that Rory did a deal, I think, in the last week, or scaled it. I really liked it personally, even though I didn't examine it. You did sort of an AI for wealth management or asset management, right? I love this for a lot of reasons. I have some questions, but Google isn't going to copy that. They're not going to copy automating trust and estate planning, investment advice, tax efficiency, or your investment legacy planning. Maybe it doesn't do all of that. Those are spaces where you have incumbents, but maybe you have some space to run, right?

Agreed. The model provider is not the constraint there. And, for the record, I would say I don't think the model provider will be the competition in many apps. Going back to where I started, I think the OpenAI code red this morning was frankly tantamount to an admission that we need to do our core mission for the next year, and probably less futzing around in other things. That sound you might hear is the consumer hardware product slipping out.

I do believe that more of these apps are defensible. I think the model providers will be there. I think coding is obvious. But even when you get much beyond that, I think if I were on the board of OpenAI, it would be, "Win the ChatGPT wars and you are worth $2 trillion. Let's not fuss around with little vertical markets that can be worth a couple hundred million bucks. Why are you even talking about this?"

Especially when everybody poked poor Google and Microsoft and said, "We'll make them dance." OpenAI kind of laughed at them, and now they're poking back. You put all your effort behind that. We may have seen that the models are going to do everything. I mean, they're going to do coding, but I don't know if they're going to expand into all these verticals at that level, right?

And then, yeah, thank you, Jason. On the Range wealth management, I think the interesting comment—and again, we're always loath to just push our investments—but I think the big-picture story there is: can you use AI to automate not just selling software to wealth managers, but the business of wealth management? And this is the key sentence that I like, because I hate the word wealth management.

The idea is you can go much further down the wealth continuum and give the same kind of product that the super-rich get in terms of managing your stuff and managing your taxes, which, as Harry knows, in the UK are now north of 50% and getting higher. So, you want to be able to manage your affairs, file your taxes, and there's a whole ton of that work that's done expensively with humans that can be done really cheaply with AI because it really is just: follow the law, fill in the forms, do the work.

Hopefully, the idea there is you automate a lot of that, and then you can deliver a high-quality product to a much broader marketplace. One of the big-picture things I think that's always true in investing is, whenever you see a product that only really rich people have, if you can find a way to get that in the hands of the rest of us, we all want it too, right?

Yeah. The only thing is, I wondered—sorry, I didn't mean to harp just on what do we call it?—the terrible TAM.

No, the TAM trap.

The TAM trap on wealth management. What's the vendor called? Sorry, I should know.

Range. Range. Sorry, there was a little echo.

8. The Importance of Compounding in Investments

I love it. I get the problem. Anyone who's lived it, who's gone through any of this stuff, could talk about it. But I think you do have to be smart about the TAM, right? Because they're going to charge $8,000 to $10,000 for something that you pay a bunch of numbnuts $30,000, $40,000, $50,000 a year for if you're wealthy, right? Maybe more.

But it's not 10 times the price of the existing product, right? You've got to be smart, because you can pretend everyone in the world will pay you $10,000, but you also have to be rational to not have a TAM trap, right? Because Wealthfront's trying to go public, right? In theory, maybe it's a comp. In theory, Wealthfront should be a $10 trillion company. I mean, everyone could use this product, right? But in reality, there is some TAM limitation for Wealthfront.

I totally agree. It's all about segment. I love Wealthfront as a comment here. Wealthfront and Betterment, I really love those companies because, again, it's back to the same thing. I get uncomfortable in the wealth discussion because who gives a damn what the ultra-wealthy have to deal with, right? What I loved about Wealthfront—what I like about these—is Wealthfront was saying that even paying 50–70 bps to someone to manage your money is crazy, because we can just put it in this automatic thing and do it automatically for 10 bps.

The thing about those businesses, to your TAM comment, is they actually take a long time to build because the whole value proposition is we're charging you less and you just get to compound more. But in the end, they're lovely businesses, and we looked at—actually, not Wealthfront but Betterment—10 years ago, and we figured it would take about this long, 10 years, to build. Because remember, if you're charging 1% of assets, a billion is a lot of money. If you're only charging one-tenth of that, you need $10 billion to get to the same place, and if you're targeting people with less money, by definition it takes longer.

So, these businesses take a long time to build, but I think when they do, they're way more powerful than some quote-unquote wealth manager that's really good because he takes you golfing and gives you a PowerPoint once a quarter about how badly your money is doing and how they're really sorry.

So, I like that. Or even better, they ask me if I want exposure to private equity and venture. That's the main value I get from Morgan Stanley.

“Hey, Jason, it's your adviser this quarter. I can get you into a hot venture fund you've never heard of. The returns are negative at the moment, but it's a lengthy J-curve and it's a hot deal.”

It's a 3-and-30, triple-layered SPV. Have you ever looked at my account? Do you know anything about me?

Sorry, I didn't mean to be divisive. I didn't like this deal. I saw this and I thought, gosh, sorry.

I didn't see it. Why didn't you like it?

Why? What big business, and what good business, really big, has been built in the wealth management space? Wealthfront.

Merrill.

What?

Just pause, pause, pause. Okay, one very old one, but we're looking at Wealthfront. What? 17 years in existence and every good investor in there. How big is that in the opportunity-cost world that we live in?

But the important question is, why does—okay, listen. There's a gap. There's Wealthfront and Vanguard at the bottom. I want to hear Rory's thoughts. And then there's the world's crappiest product, which is Goldman Sachs and Morgan Stanley, which take 1% of your assets and do nothing I can see other than give you loans, which are very valuable, right?

For folks that don't know, if you're sitting on $50 million of NVIDIA stock right now, you can sell it and pay $25 million in tax, or a bank will give you a loan. Now, it's not cheap today. It's 6%, but that's a lot better than 50% if you can deploy it.

That's the only product I know. They'll tell you they'll help you with your trusts. They don't. They refer you to someone who doesn't call you back. They'll tell you they'll help with your taxes, and they'll tell you, “We're not allowed to talk about taxes.” So, they can't really do anything except give you a loan.

There is such a gap in the middle. If AI can do estate planning, taxes, all this, then something that was crummy, to Harry's point, might become great. If AI lets you do it, right, it might become great.

Rory O’Driscoll

That's exactly right. First of all, you just nailed it. That's exactly the value proposition. I like it. You deliver trusts, you deliver estates, you deliver all taxes, because when you have to file your taxes, the fact that you do your estate planning with one person, your taxes with another person, and your wealth management with a third party is absurd. It should all be under one roof.

But I want to go back to your comment, Harry, on Wealthfront, because, again, I'm always uneasy just pushing our company. Let's talk about an excellent company that's not ours. Yes, it's taken a long time to compound to here, right? But not everything is tech-first, where the adoption cycles are 5 years.

I think when you launch a company like Wealthfront, you know the adoption cycle of something like that is going to be 10 or 15 years. In my view, it's just on track, and the financials are lovely because, at scale, asset management is a wonderful business. They are providing a cheaper product than anyone else at 10 bps, and they've lined up a bunch of millennials and whatever the generation a little bit older—30-year-olds and 40-year-olds.

Over the next 10 or 15 years, that generation is going to get rich. They're going to get rich with Wealthfront, they're going to keep their money there, and it's going to be a compounding machine, just like Charles Schwab was a compounding machine when they started in the '70s with cheap brokerage. It took a long time. They were actually owned by Bank of America and then spun out in the '80s—I was there.

In the end, it just compounds because, over time, the great thing about it is that it does compound. Early on, wealth management in any form is a tough business because it takes a long time to build, but when it does build, Wealthfront is over that gap now. It's going to be there for the next 30 years and, frankly, in a way that a lot of pure tech companies won't.

I get you on the compounding machine, and I share your view on the beauty of those businesses, but actually, you're competing for dollars against the same people: your Kleiner Perkins or your Andreessen Horowitz, where LPs can put money in their funds or your funds at a B2B stage, and they are in Glean and Rippling and in the race to $100 million faster than we've ever seen.

Those LPs will be going, “Well, those firms, they're more exciting,” and you're going, “Oh, but it's compounding. It's Charles Schwab 2.0 coming. I promise you. Watch the pod.”

I don't know that that's 100% true, though, Harry. I think it's 80% true, right? I think if you're going in to raise capital for your fund and you're being compared to Glean and everything else, but, I mean, Rory has more experience. I think as long as you have the numbers, LPs are kind of excited if you have a slightly different way to get there.

Slightly—but you've got to have top-10% numbers. Do you think so? If you've got companies that are kind of—and, actually, to your point and to our point, bluntly, growth rates are so much higher and growth expectations are so much higher. If it is a slower compounder, the next round is less certain than ever. It's less guaranteed.

Yeah, but LPs are looking backwards. If you're sitting on multiple high-performing funds, you're going to get a fair amount of flexibility today.

Rory O’Driscoll

I think the point is that it's hard not to be seduced by the hottest deals today. And, by the way, LPs love to see great follow-on investors. Sequoia came into this, and Andreessen—

But I want to push back a little. I do take on board your point, and there's no doubt that the velocity of validation is super strong for AI companies right now. If you want to do a deal with the highest probability of a step-up in the next 6 to 9 months, you should do an AI company that's raised at a $1 billion pre, because 40% of the unicorns in Q1—I said 23% last week, and one of my colleagues corrected me: 40%—of the unicorns that raised in Q1 as a unicorn for the first time have already had a follow-on round.

Rory O’Driscoll

So I think, Harry, you’re exactly right. If you want to buy short-term momentum, that’s a great place to play. And it’s not just short-term momentum; it’s also driven by great performance. So yes, that’s absolutely a good slug of what you’re doing.

But in the end, the biggest uncertainty is not, “Can you get a markup that’s nice?” The biggest uncertainty is, “Can you build a big company here or not?” And there are so few times when you can say, “I believe you can build a big company here,” that you shouldn’t then screen out and say, “Oh, I can build a big company, but it might take a little too long.”

Because there’s a rule in engineering that you’re only as accurate as your least accurate variable. In other words, if you have 6 or 7 variables that go into something, your accuracy is determined by the thing you know has the widest variance. And if you have high certainty that something can be a company, that’s the hard thing to do. If you’ve got that and everything else, you can adjust for valuation, you can adjust for time, et cetera.

I would love to be in Wealthfront, for example. I think it’s just an awesome company. I think that will compound, and you’ll hold it, and 15 or 20 years from now—it’s going to sound pejorative to AI. It’s not. I love that space. It’s where I play most of the time.

Harry, I think Charles Schwab went public in either 1982 or 1983. It’s public today. It’s worth $60–80 billion. Name me 5 tech companies that went public in 1983.

Dude, are you kidding me? I was born in 1996.

Rory O’Driscoll

But my point is this: tech companies, they come quick and most of them go quick. Now, by the way, if I’d said 1986, you could have come back to me and said, “Microsoft, Oracle, and Adobe.” If I’d said 1982, you could have said Apple. That’s why I think I picked 1983.

The point is that these singular, different companies—these companies that are off the beaten track—often take longer to compound, but they end up with more empty space. And as I say, Schwab has compounded for 3 or 4 decades, and God knows where you are in it.

I get it, but sorry, I didn’t mean—I just think we’re playing a relevance game, and I think this is the honest truth about new-age venture. We’re playing a relevance game where Ramp raises 4 rounds in a year, where media matters more than ever before. Duh. All of us here. And where you’re like, “Ah, it’s slow compounding coming soon.”

Rory O’Driscoll

So slow, by the way.

It’s just a tougher game. And I think LPs are seduced by incredible follow-on investors, quick up-rounds, and numbers still. And I’d rather be playing that game than the “it’s coming” game.

Rory O’Driscoll

And you’re right, Harry. You should have done that, provided you’re also right about the underlying investments. If you play that game and you’re wrong about the investments, then you’ll just be the guy who did a load of high-priced rounds in a deal that didn’t work, right?

So I agree with you. But again, I go back to my comment: when you have high certainty that a big company can be built here, you weight that more highly than everything else.

I think, actually, Peter Thiel—as with all intelligent venture comments, when you go back long enough, you discover Peter Thiel made them already. I think he said somewhere in his book something to the effect of, “All that matters is, can you build a big company here?” And literally he said, because that rule is so hard, because it’s so hard to find them, they have no other rules. Their perspective is, once I filter for that, I can’t have any other rules on stage or sector. I just want big.

That’s what gave them the courage to do biotech, defense, and space. We’re not as brilliant as that, obviously, but I think it’s some version of that rule, which is: when you see a company that can be big and you see it’s tracking to be big, prioritize that over hype and FOMO.

Rory, can I be absolutely savage? Do you have to be in that slow-compounding-picking strategy? Because there are 2 worlds in venture. There’s the obvious and really competitive: insane growth, really, really obvious. And then there’s, “I’m going to be smarter, pick the compounder, see beauty where others don’t.” Do you have to be here, respectfully, because you’re sitting in the Valley at Series B and you’re against Andreessen, Founders Fund, Sequoia, and you can’t beat them?

Rory O’Driscoll

I think you have to do both, and you can do both. Again, I’m not sitting here going, “I want to filter for X, Y, or Z.” I want to filter for great companies. Then I have to win them.

And you’re right. If you’re identifying a great AI company in XYZ space, then you’re going to find way more competition, which means either you’ll lose or you’ll win and you’ll pay the market price to win, which won’t be cheap, right? And that’s one way to make money. That’s most of what we do.

But you can also go and look where I go, “Oh, I think this is interesting and differentiated.” And as long as I have the same conviction on the ultimate outcome, you can do both, right?

So you’re trying to apply a momentum and hotness rule, and I’m trying to apply a “will there be a big company in the end?” rule. And I get the interim consequences. To be very clear, someone said it to me 20 years ago as an LP, right? He said, “There’s no such thing as blue-collar venture.” It was a brutal comment, but I think it’s your point, Harry, right? There’s no such thing as randomly non-cool stuff.

At the end of the day, we’re building high-growth companies, and you’re not going to make it on value. You’re not trying to choose on value. You’re trying to choose on certainty of a big outcome.

For what it’s worth, listen, I think it might be a fool’s errand to invest in things that you’re just very interested in. But I think there is an advantage to it. And I’ll tell you what I’m interested in for 2026 and 2027.

This is why I like Rory’s investment. I know this sounds obvious, but AI for coding is great, but we didn’t even figure that out. Claude figured that out. Cursor didn’t figure this out. Replit and Lovable both didn’t figure it out. I can tell you the story: Claude figured it out. Anthropic—the guys, once they quit OpenAI, they figured it out—and everyone grafted on this, including Gamma.

What I like is the next generation. Can AI take large markets like wealth management that don’t work today? Can AI really, for real, with Claude Code and everything, utterly disrupt it? And I think that can be huge.

I’ll give you an example. I set up 3 trusts, okay? The wealth management didn’t help at all. And then I went to the lawyers, okay? And it took me 11 months to set up 3 trusts. And I said, “I’m really frustrated this took too long,” to this guy who’s a celebrated trust lawyer in Silicon Valley. He’s like, “Well, good news: most of my clients never even finish them.”

I’m not saying momentum investing isn’t the right thing today. If you can come in with AI and magically take every single frustrating part out of it—and the Wall Street Journal just said today that the average American retiring has like $1.8 million in cash and equity—if you can take all the friction out of that, all the friction out of retirement, wealth management, investing, trusts, redeploying QSBS, and everything because of AI, I think you could build a $20, $40, $50 billion company, and I would at least want to take the meeting.

I’m just interested intellectually. Can AI solve some of the biggest headaches that we see that maybe you don’t see every day? Maybe they’re in environmental compliance. Maybe they’re in other things. But it’s at least worth a meeting to see if they can utterly disrupt how it’s done, right?

Rory O’Driscoll

Yeah, I really like the pitch here, which is: it’s not the ultra-wealthy. They’ve got a million people who flatter them and do their work for them and charge them a gazillion dollars. It’s the small entrepreneur. It’s the doctor. It’s the dentist who’s earning good coin, because we discussed earlier that dentists and doctors are well paid, but their affairs are modestly complex.

They don’t want to screw up their Roth IRA withdrawal. They want to leave their house to the kids. They don’t want to have a big estate-tax problem. Those are the kind of things more complex than nothing, but not where you can spend $20,000 on a lawyer to fix it. That, I think, is a huge market.

I’m sure people in the UK know this, but, in fact, in the US, unlike the UK, everyone has to file their own taxes, which is just— I mean, I’ve paid taxes in Ireland, England, and the States, so I know the different systems. Everyone has to file this god-awful tax return every year, and the minute your affairs get even mildly complex—you do a rental property, you have a distribution, you have a capital gain because of an investment you made in a restaurant—suddenly your tax affairs are complex, and if you screw it up, you end up paying more money.

And those are the people who need that kind of mass-market wealth advice on how to handle their affairs better. And right now, it’s a disparate group of attorneys and accountants, and it’s quite messy.

I do think AI, like a lot of other markets, is a little like health care that your prior speaker was talking about. There’s a whole bunch of this knowable but nonetheless complex information that has to be assembled and marshaled. And at the right time, the person who has the question has to get the right answer.

That question can be, “What form of cancer is this based on the CT scan?” Or, “How much do I owe to the US government based on these facts?” But you want the right answer at the right time.

And it turns out machines do that a lot better than humans.

My takeaway is you need a new wealth manager, dude. Seriously, a Goldman not just doing the gold.

Harry, one comment here: I believe you.

Rory O’Driscoll

Not everyone has a wealth management problem. Harry, how about we focus back on the problems of the people—the founders who are trying to build wealth—rather than you as a rich guy trying to spend it?

I’m a humble podcaster. Rory, I believe that AI will disrupt some of these categories and build huge businesses. I believe it. Maybe be capital efficient until you prove it. If Rory’s deal does $100 million growing 150% or 200%, everyone will flood into this.

Absolutely. I would say it made no sense back in the day. So, not every founder can directly compete with Cursor. As founders, we have to play to our strengths, and if you find someone like that, I might have invested just because I’m passionate. Maybe it was a bad idea. I mean, I’m into it, but you might have convinced me. Maybe just don’t spend all of it.

Rory O’Driscoll

Maybe a little more conservative.

Did he not bring you into the deal?

Did not bring me into the deal.

Oh, he brought me into the deal.

Rory O’Driscoll

Yeah, did not bring me into the deal. I think—shut up, Harry—because Jason’s made a really important point that I think is relevant to our wider founders.

I think, Jason, seriously, your discipline as a founder on capital is exactly proportional, in part, to how hot the market is perceived to be. And if you’re playing a game in a market where you need to make progress before you can raise, then you don’t have the ramp for a raise-before-prove-it strategy. You have the capital discipline to prove your point.

In the end, if you prove it, you’ll get the capital. But I think, Jason, that was frankly a spot-on comment. We could send you as a board member. Literally, it’s like, “Here, guys, a little value. If you prove it, the world will be beating a path to your door. But if you haven’t proved it, you’ll be screwed.” They’ll get it done.

9. Supabase at $5BN or Lovable at $6BN: Which One?

I think that’s a great note to close on. Guys, thank you so much for being part of a live show.

There’s no Khosla quickfire.

There. You want a Khosla quickfire?

Rory O’Driscoll

The one quickfire is Supabase is now at $5 billion. Lovable is at $6 billion. Now, obviously, Lovable uses Supabase for every instance. Would you rather be in Supabase, or would you rather be in Lovable? I’ll say Lovable.

And I’ll tell you why. I think Supabase is benefiting from the trend of vibe coding, and Lovable is fundamentally a bet on the front end monetizing on vibe coding. One of 2 things happens: either vibe coding is a category, or it’s not. If it’s not a category, both of them are screwed. If it is a category, Lovable gets more of the money than Supabase, just because it’s the front end and it gets, I don’t know, $20, and they pay $2 of it to Supabase—some version like that.

So, if you’re in a highly risky category, the dumb bet is to say, “If I win, I get a little, but if I lose, I lose 100%.” You may as well be in for a penny versus in for a pound, as we would say in the UK. And the thing about the Lovable bet is, if you win, you’re going to win big. Which, of course, is why a smart young man like you is in Lovable, Harry.

See, I thought I put in 1 pitch for you there, dude.

Thank you so much.

Rory O’Driscoll

I would take Supabase.

What?

Rory O’Driscoll

I would take Supabase.

No, I’ll tell you why. For what it’s worth, this is how I’m feeling today because of stability. I think Supabase is a harder problem to solve.

Rory O’Driscoll

Yeah.

And so, right now, this is where I’m conservative. There’s just so much change. I don’t know what Google is going to do. I would just prefer a harder problem today. Even if my returns were the same or lower, I’m sleeping fine, but I’m just anxious about cloneable stuff, and I want hard problems. They’re reassuring.

At the end of the day, I think it’s just a fork of Postgres. It’s open source that they’ve redone. It can be done again. Neon did it, and Databricks bought them for $1 billion. But databases are a hard problem. You can only lose so much data. You can only have so many issues. You have to figure this out.

And 5 years of investing in a database that everybody uses—it ain’t so easy to churn and leave your database, right? This is what I’m thinking going into next year, if we want to close. This was a year where we tolerated a lot of churn. We only cared about growth. This was the year of growth, but nothing else, and so is next year.

But I’d love a little defensibility. I just love a few hard, freaking problems. We’re past the thin-wrapper layer, but I’m going, “What are you picking between Lovable and Supabase?”

Rory O’Driscoll

Oh dear, Lovable all the way.

Lovable all the way. We’d love to lead the seed.

Rory O’Driscoll

Loyal to his paycheck.

Yeah. The loveliest thing about this is we knew each other well before, but the friendship that we have as 3 now, having done this show, is just freaking awesome. Honestly, it’s one of the highlights of my week doing this show every week. I’ve never said this to both of you: I so appreciate the friendship that we have, and thank you for doing the show with me because it’s always so much fun.

I learn so much, and I get so many messages from founders who learn so much. So, thank you for putting up with me, both of you. I know it’s not always easy, but you’re awesome. And thanks for everyone who stuck it out to the end.

Rory O’Driscoll

Appreciate it.

Take care, guys.