Gokul Rajaram谈公司需要的8道护城河，以及辍学生为何正在“AI Maxing”全世界

I call it the eight modes: data mode, workflow mode, regulatory mode, and distribution mode.

We're on number 5. I'm loving this.

Ecosystem moat, network moat, physical infrastructure moat, and the 8th one, I would say, is scale moat. I think anything 4 or more, you're pretty damn secure.

I'm thrilled to welcome one of the best operated turned ambassadors of the last two decades, Gokul Rajaram.

You cannot be a single product company. Vertical products, you've got to really own full stack. It's harder otherwise to be a 10 plus billion dollar company.

What's the biggest miss? Is it Braintree?

Most recently Quince, but to be honest, even bigger miss than that in some ways. It's not a miss in terms of investing. It's that I didn't get that.

Gokul, I've wanted to do this for years, and you've clearly played hard to get, to put it mildly, after my continuous WhatsApp messages. Thank you so much for joining me, man.

It's my pleasure to be here, my friend. Thank you again.

I wanted to start with how some of the prior companies you've worked at have shaped your investing mind-set specifically. I wanted to start with Google. When you reflect on your time with Google, how did that shape your mind-set for the types of companies that you like today?

I think the best way to think about my Google experience is that Google taught me that, ultimately, the best companies have a remarkable product at their core. Google was a remarkable product. Google definitely had a philosophy of, “Build it remarkably, and they will come.” Go-to-market was not Google's specialty, but what Google was really good at was building an amazing product. Sometimes the go-to-market worked; sometimes it didn't work. At the core was a remarkable product.

Ultimately, my core investing thesis is that if there is not a remarkable product, all the go-to-market and distribution in the world will not save you. I look for what the remarkability is in the core product or value proposition of the company. Is it 10x or 100x better than the alternative?

I'll tell you a story from Google. When I joined in 2003, there was a project going on called Caribou internally. I was like, “What the hell is this?” This was webmail, which gave you 1 GB of free storage. Back then, Yahoo Mail offered 10 MB of storage, so it was 100x. I thought, “There's no way this is possible,” and it turned out it was. It was released on, if you remember, April 1, 2003, and people thought it was an April Fools' joke. But that was Google literally taking something that was unbelievable and making it a reality. That's the kind of product I like: something remarkable, something unique, something powerful.

I like it. It reminds me of Neil Mehta, who talks about jaw-dropping customer experience as one of his core monikers for thinking about companies and investments.

Next, we have Facebook. How did Facebook impact the types of companies that you like?

What's interesting is that even if you have a remarkable product, you still need distribution. Facebook taught me the power of distribution. Mark Zuckerberg is probably the best distribution genius in the world. He would look at a product and say, “This is how this product is not going to work.”

It taught me the power of multiplayer products in particular. I think most software products are single-player, and as soon as you make them multiplayer, there is a uniqueness in switching, distribution, and so on that comes about. Facebook, by nature, you can't use it if you only have 1 person on Facebook.

When I saw Figma, the power of Figma, I felt, was not just that 1 person could use it, but that it was much easier to share with other people in your company. I think the best PLG software companies are those that multiple people can use, because it increases defensibility. So, the power of distribution and multiplayer products.

What about Square?

Square was an amazing journey.

What did you learn from Square that you've taken to your investing?

The power of a multiproduct portfolio. At Square, when I joined, we were a single-product company—payments and payments only. When I left, we had, I think, 11 products, each doing more than $50 million in revenue.

One of the interesting metrics was that we moved our key North Star metric to the median number of products used by a seller, by a merchant. It turns out that the more products a merchant uses, the more retentive they are and the stickier they get.

This is the other thesis I have, and it's obviously very clear now: You cannot be a single-product company. You've got to make sure, most importantly, that product 2 emanates very naturally. It can't be this completely separate product. It has to be very adjacent to product 1.

For Square, it was a product called Square Capital, which was basically a cash-advance product that came from the fact that Square controlled the payment flows and knew exactly the merchant's credit history. It could underwrite based on money going in and out, and it was a beautiful product.

The interesting thing about having a multiproduct portfolio is that not every product needs to generate a profit. People always say, “Oh, it's not making money.” Square Capital didn't make much money, but it was very good for retention. Some products are good for making money and are part of the profit pool, and some are good for retention. Companies need to be very clear about which are the profit-pool products and which are the retentive products. If you confuse the 2, your teams don't know, and they build for the wrong outcomes.

The power of a multiproduct portfolio is being able to have products with different goals: retention versus profits.

I love that it doesn't need to be profitable. I also see so many investors today being relatively inelastic in terms of their mind-set on margin. Whereas I hear, “The margins are [__],” and it's like, I didn't go into the next company with negative gross margins for the first year, I think. Negative gross margins.

Dude, I don't think DoorDash had great gross margins for the first years either. I don't think Deliveroo did, which DoorDash obviously acquired. It's funny that we repeat the same mental cycles: “Oh, the margins are shit.” And it's like, yeah, so were the best companies' margins. Spotify didn't have great margins for a very long time, and its margin increase has been amazing, actually.

What is the lesson from DoorDash?

DoorDash was the most operational of all 4 companies. I thought I was a good operator. When I got to DoorDash, I really realized what operations means. A lot of my philosophies around how to truly operate in hard mode have been shaped by it.

It really was the apotheosis, or the epitome, of how product and operations can work together in the physical world. How it shaped my investing philosophy is through the kinds of people who came out of DoorDash. I think they're excellent, and I try to get them. It's really around hiring, talent, and taking really hard problems and solving them.

I'll never forget when COVID hit. As you know, most restaurants were shut down for the first couple of weeks, and DoorDash had to make a very hard call about what to do to get these restaurants to open. Ultimately, we decided not to take any revenue share from these restaurants for a month. Even though we were a private company, we had a small amount of cash on the balance sheet, and that really hurt. It was the right thing to do in the long term, but it was extremely painful in the short term.

You spoke about the skill of operators to work both in a physical and in a software-based environment. Now, with DoorDash and with Project Europe, which we chatted about before, we're seeing all of our hardware companies be so freaking popular right now because everyone's terrified that, bluntly, Anthropic is going to eat their lunch, as we keep seeing with Anthropic doing security and security stocks plunging.

I want to talk about the SaaS apocalypse, because my job with this show is to learn from people much smaller than me, and I'm lucky to do that here. Is the volatility that we're seeing justified, or are we in a manic, hype-and-oversell environment with emotional volatility?

Well, all of our software portfolios are deep red, right? All of us have some software stocks, and the reality is that the public market has decided that since code is becoming free at the lower end and becoming much easier to generate and create at the high end, every software company is going to zero.

I think this is a 100% overreaction, because not all software companies are created equal, and we can talk about what the differences are. I actually spent the last few days thinking about the different characteristics of a durable software company. Both you and I think about this a lot.

What are the characteristics of a durable software company? I think there are a few that we can talk about. But, yeah, I think it's absolutely an overreaction. Everything has been painted with the same brush at this point. It is absolutely an overreaction.

You're going to leave me with a cliffhanger, Gokul. You're like, “There are some very durable characteristics. We can talk about them if we want.” I would love it if we could talk about them. Can you please help me understand?

It's basically a play on Hamilton Helmer's 7 Powers, but it's slightly different. I call it the 8 moats. The first moat is the data moat, which we all talk about, but it truly has to be proprietary. It has to be data that nobody else has access to. I think Spotify is a good example.

If you look at their discover product, it uses a decade of listening behavior across hundreds of billions of people. You can't create that discover product easily. The second is the workflow moat, which a lot of people argue is a weak moat. I agree that by itself it's a weak moat, but the deeper you're embedded in the company—running their operations, moving their money—the deeper the workflow moat is.

By itself, I don't think it's enough in perpetuity, but the deeper your embedding is, the stronger it becomes. For example, NetSuite is an ERP that runs your business. They have a much, much deeper moat than, say, Zendesk, which is a lighter workflow moat. You could say NetSuite is a 1, while maybe Zendesk is 0.5.

The third one is a regulatory moat: licenses, capital required, and multi-year procurement contracts. Coinbase, where I'm on the board, is a great example. They have MTLs, or money transmission licenses, state by state. They register with FinCEN, and all of those things make it impossible for a company to use anybody other than Coinbase to custody its crypto.

The fourth moat is a distribution moat, where you have proprietary, exclusive distribution. Intuit is a great example. Anybody who wants to build an accounting system should think about it. When I started a company after Google, I basically tried to use this company called Xero. I was like, “Let's use Xero. It's the new thing. It had just started, and it seems like a cooler interface.”

My accountant said, “No, I'm sorry, I don't use Xero.” He shut it down. I had to cancel Xero and go to QuickBooks. What a great distribution moat there. You've trained a network of CPAs to only run QuickBooks. I don't know if they have a commission or what they get, but that's a proprietary distribution channel that these guys have. It's very hard to displace them.

The fifth is an ecosystem moat. If you have a platform or ecosystem that many third parties have built on and rely on, you have a moat. Shopify is a great example. You can vibe-code an e-commerce hosting platform, no problem. But can you vibe-code the hundreds of thousands of developers and third parties who built all these applications on Shopify?

Every Shopify merchant I know uses at least 5 or 6 other third-party apps. That's a huge part of the Shopify ecosystem. That's a moat.

The sixth one is a network moat. That's classic DoorDash. I think DoorDash has many other moats, but AI can vibe-code the ability to access restaurants. It can't vibe-code liquidity, courier density, reputation history, and all of those things. Marketplace density is a network effect, which is structural.

The seventh one is the thing you mentioned: physical infrastructure, atoms. Wherever you have atoms, it makes for a moat that's hard to displace. Again, I think humanoid robots will maybe start taking over at some point, but it's probably a few years away.

The eighth one, I would say, is a scale moat. If, by virtue of your scale, your costs are so low that it's hard to replicate, that's a scale moat. I think Amazon is a great example. TSMC in semiconductors is another example of a scale moat.

Those are the 8 moats: data, workflow, regulatory, distribution, ecosystem, network, physical, and scale. What you do, I think, is take a company and score it across them. Maybe you assign 1 point to each moat they have. Anything at 4 or more is pretty damn secure. If you have 2 or 3, it's a weak moat. If you're at 1 or less, you probably need to build some more moats, or you need to do something to make up for it. If you have 0, you're screwed, basically. I'm just saying this as a thought I had.

We have Atlassian and Monday. They're both down about 75%. I've had both their CEOs on the show. If you look at them and put them across these 8 rules, you would probably say that Atlassian is being massively oversold, and that Monday, as awful as it sounds, is maybe being rightly priced in this environment.

This is Sean Ellis. How would you think about Klaviyo in this way? When you look at the ability for public companies to build good agent products, it would seem very obvious that Shopify will build Klaviyo now in the need to reaccelerate. How would they rate?

I don't think Shopify will build it. Shopify is an investor, and Shopify, I think, has decided—at least in my opinion—that this is not part of their mission to build this product. So I don't think their risk is Shopify. It is that it has become easier to build Klaviyo now than it was a year ago. It's easy to build Klaviyo.

They do have—I haven't talked about brand. I think brand is no longer a strong moat. I explicitly excluded brand. I don't know how strong Shopify's promotion of Klaviyo is. I think a lot of it depends on how strong and tight the proprietary distribution they get from Shopify is.

If Shopify is actually going to promote them—when you search for messaging or communications, if they are a preferred product and they have a relationship that makes that work—I think it's very hard to displace them. It's hard to at least displace that part of the business.

Dude, you just threw a grenade in there. Don't expect me not to pick up on it. I just had Lena Varnay, who's the head of growth at Lovable, on our 20Growth show, and she said that brand is actually the most important thing. As you commoditize technology and it's easier and easier to create, how people resonate with a brand is the most important thing. Why do you think brand moat is not as important?

Businesses are much more rational in thinking about it—less irrational, I think—and the alternatives are going to be much stronger. On the consumer side, consumers are much more like dollars and cents. There are dollars and cents in business, too, but there is a natural inclination to just trust brands. I think on the business side, it is going to get weaker.

I actually disagree a little bit, because switching costs are so much lower. One of Hamilton Helmer's 7 Powers is switching costs. I think switching costs are going to go essentially to 0, because over the next 1 or 2 years, the ability to port your data as a business or consumer from any ecosystem to another ecosystem is going to be very easy.

People are going to be able to replicate almost pixel by pixel the experience you have with one product in a different product. You'll have clones popping up left, right, and center, and data portability is going to be easy. In that case, what is that brand really? It's like in professional sports: you kind of cheer for the player or the team when they switch teams?

I need your help, because the one that I continuously oscillate on is Salesforce. When you talk about data portability becoming increasingly easy, we had Zapier from Cloudera on the show. He said agents would make data migration from systems of record increasingly easy, so they wouldn't have the lock-in that we think about, or reduce switching costs. But then I look at your 8 factors and I'm like, well, they have workflow, they have distribution, they have ecosystem, and they have scale.

They don't have scale. Their scale means that it's cheaper for them to produce software. I think software earlier was a scale game where, because you had to produce a lot of software, it was cheaper for you to produce a lot of software. Guess what? Now everybody can produce software as cheaply as anybody else.

If they had their own data centers, like the hyperscalers, I think the hyperscalers are the ones that are able to say confidently that they have scale, or people in the physical world. A pure software company can't get that scale moat. But yes, they are very similar to Atlassian, where they have a score of 3, I would say.

So do you think Salesforce and systems of record like Salesforce are inherently attractive or less attractive, given that data portability is increasing?

They are more attractive than most software companies. But if they don't build agentic workflows and commoditize a complement by figuring out where the profit pool is, I think they have to figure out whether the profit pool is in the data or the workflows.

If it's in the workflows, they need to make data storage free and basically change pricing to an outcome-based model based on workflows.

If they feel the profit pool is in the data, then they need to give away these workflows for free. They need to commoditize all the agentic companies that you and I know are trying to build on top of them and charge for that. They need to build better products using their data and make them free. I think that's the way a NetSuite or a Salesforce system of record needs to operate. They have to commoditize the complement. They can't just wait around for other people to build on top of them.

We're seeing buybacks like never before from your Workdays and your Salesforces of the world. Is that truly indicative of internal company confidence, or do you think it's a necessity to externally show the world that they are confident?

It's both. They're confident internally, but they need a signal to show that they are confident. I think the founder buyback is the strongest signal. It's not just the company, but also the founder. Did you see the ServiceNow CEO's $3 million buyback? Then they saw that his garage of classic cars was about 3 times as much. I was just like, “Oh, that's a bad comms move.”

Yeah, I think there are buybacks, and there are buybacks with a capital B. You want the capital-B one. You want a CEO of a large company to do a $20–$50 million buyback to show confidence.

I agree.

We've mentioned Monday. We see companies like Notion and Amplitude. I'm mixing public and private companies, but growth-stage companies and even private companies do bolt-on strategies: “Hey, our core product—we're the bolt-on of AI.” How do we determine bolt-on AI strategies that work versus bolt-on AI that doesn't?

I think the bolt-on AI strategy by itself has a real ceiling, but I think the companies where the bolt-on really works are the ones that reframe what the product does, not just add the capability. For example, if you just add AI search, that's one thing. If you build search as an experience with new UX primitives, that's completely different. I think one is just an upgrade, while the other is doing something completely different.

I think very highly of Notion. Notion is adding AI, as you mentioned. They're adding a lot of AI agents, and I'm hoping that the way they've added it continues to improve. I actually played around with the product. I think it's pretty good, but the AI agents now need to get better based on how users interact with them, and they need to tune the model for their customer base.

Most bolt-on players aren't doing that. They're simply using a GPT or Anthropic model and basically just adding a thin layer. You have to rebuild the entire experience end to end.

Do you do that by identifying something where AI doesn't just improve the margin, but changes the experience and the economics?

Document processing was a good example. Until about 6 or 9 months ago, you couldn't actually extract structured information from unstructured documents. Now you can reliably process dense legal contracts. Your experience around documents needs to be fundamentally different.

If you're just getting someone to upload a document in a flow, you need to instantly give them insight from the document while they're uploading it, versus having the same document-upload experience. That's crazy, because now you should be able to infer, for any document that enters, what the hell is going on instantly. You need to reevaluate every single interaction and see what has changed.

That's the biggest difference in product development today. Model capabilities are improving every 6 months, because if you have too long a product roadmap, you're going to run into a problem: the next model comes along and blows it out of the water. You've got to really understand the capabilities of each new generation and make sure you don't have too long a roadmap, because your roadmap is going to be blown out by the next model iteration.

Speaking of being blown out by the next model iteration, how do you, as an investor, educate me? How do you ascertain safety from model intrusion versus being in the way of models and being eaten with the next update? If you have some of the other moats that are physical and so on, it becomes easier. If you're a pure software company, which of those apply to you?

I think fintech is a good one. Fintech goes through these cycles. I think fintech, especially at Marathon, we invest a lot in fintech, and we actually think, “Oh my God, fintech is one of the best models.” If you're moving money, you're generally in a good place. Anything that touches money, we feel, has a very strong moat there and is much more defensible.

Data and workflow moats are the 2 things you're really hanging your hat on as a software investor. If you're not doing fintech, at the early stage it's too hard to know what a distribution moat is unless they have some hack, and these hacks never really stand the test of time. Ecosystem is too early to say. Network effects are too early to say. Physical infrastructure—they don't have any. Software company scale, they don't have any.

So it's really about going deep into what the data asset is that you're creating. Does it get better with time? Do I believe it gets better with time? Are you building your own model over time? Are you fine-tuning a model and improving it over time? Then, how deeply are you truly embedding the workflow? Are you just a lightweight thing that the underlying system of record could create, or are you building something more? These are the 2 things that you have to hang your hat on: a data asset that gets better with every interaction, and a workflow.

It's hard, man. I think pure software companies are hard. I think application software companies—you've got to really believe that the founders can ship with great velocity to basically build that and see proof of that compounding.

I got in trouble in the partnership the other day. I'm quite grumpy, generally speaking, but I was—

No, no.

Yeah, I know. My Twitter's getting grumpier and grumpier. I know. Anyway, I've met support agents and voice agents for auto manufacturers, dentists, chiropractors, and all of these different verticals. I'm like, guys, this is like OpenAI and ElevenLabs for all of these different verticals.

Would you say, “Harry, no, no, you're wrong. They're building verticalized data over time. They're able to fine-tune their own models. They are deeply embedded in workflows, and they might have distribution”? Or are you like, “Yeah, the dentist call agent is a little bit of plaster on top of a wound”?

I do think these are viable businesses. I don't think they're going to be big businesses. As soon as you go vertical, I don't think you can do one function within a vertical. I think what you want to see there is the ambition and ability to truly own the full stack and build the whole product for the vertical.

ServiceTitan, for example, is a canonical example. It went public last year, with a great outcome. It's still a sub-$10 billion company or something like that. If you look at the S-1, they have 32 different products. Even after selling 32 different products and really being, at least in the U.S., the canonical company for any service professional or physical field-services company, they still are a $10 billion company.

What's astonishing when you compare that to Robinhood is that Robinhood has 13 product lines now doing over $100 million in revenue. 13.

Yeah, Coinbase has 12 doing $100 million in revenue, exactly. With a horizontal product, you're serving a broad base. I think with vertical products, you've got to really own the full stack. I think it's harder otherwise to be a $10-plus-billion company.

I'm going for spice. In a world of 2026, can we as venture investors do vertical SaaS, given the fund sizes that we have?

I think you can. Maybe the mega-funds might say, “Look, it might not be a $100 million outcome.” But I think if you're a $200 million, $300 million, or $400 million fund, you can absolutely create a $10 million company.

Remember, one of the big changes is that vertical SaaS does take over labor. Vertical software is no longer just SaaS. It's basically software as a service, but it is services. So you're going after the services spend.

One of the interesting things, as you know, is that verticals—especially if you're selling to small businesses—spend some amount on tooling, but they spend a tremendous amount on both BPO and human capital, human labor. You need to target those 2 spends. If you do that and you're committed to building the whole product, you can absolutely do it.

You very kindly said before the show that you like the show that we do with Jason and Rory. It's very humbling when I do the show for 10 years and then find out that it's actually much more popular when I bring other people on to do it instead of me. Always good for the ego.

Rory always says to me, with AI, very simply, “We need to see the transition of spend from software budgets to human labor budgets.” If we do, the TAM obviously opens up immensely.

Do you think we will realistically see that? Maybe it's already a thing. Or do you think we will actually remain in software budgets, as we have been in some categories?

We're seeing that. I think the first one: most businesses don't want to lay off people. So the way we're seeing it, the first thing that's happening is businesses are outsourcing to third-party BPOs, many of them in India, the Philippines, and so on.

That spend is the easiest to cut because now you can offer the same service, higher quality, faster, and 20–30% cheaper. The second thing they do is, when somebody leaves, they don't replace that person. The third thing they do is lay off.

I think layoffs are still maybe a little bit of a ways away, but you're seeing absolutely BPO spend. All the call center companies that you mentioned, all the next-generation AI customer service companies, they're going after BPO budgets. I was shocked, when I was doing work in this space, by how many different verticals—doctors' offices, et cetera—use call centers outside the US.

They already have budget clearly allocated, and there's a better service. So I think it's BPO spend first, don't replace the person second, and then potentially think about laying off. Do you know Goldman Sachs and Barclays—both financial institutions—both have over 30,000 people in India?

I did not. Thirty thousand? I thought they were like a few thousand. Isn't that nuts? I was so shocked when I heard about that.

I want to understand two different types of company profiles and what happens to them. We've got private companies, and I don't want to pick on them, but it is helpful to give examples. I'm sorry, they're my friends as well, so I can kind of do it, and they'll love me hopefully regardless.

But Snyk is an amazing security business that's got great customers who love it, but it was valued at $7 billion, and it's now at $300 million ARR, growing 15%. What happens to that cohort, which is a great business serving great customers, but with 15% growth, $300 million ARR, and a very high price? What happens to that private cohort?

There are 2 outcomes for these companies. All of us have those companies in our portfolios. A bunch of them are going to become zombie companies, and they're going to try to add AI features as a last-ditch resort, not succeed, and be sold to PE.

The problem is, even that may not be a good outcome now, because PE itself is struggling to digest the companies they bought a couple of years ago, and the prices are reset. They do have good assets. I am seeing, in some verticals, companies merging with each other. We'll see if that happens just to create more scale, but it'll be interesting.

The hopefully better outcome for many of them is that, with strong leadership, you can basically burn the bridges and create a completely new, AI-native product. I think you had the Intercom person—the Intercom CEO, right? Fin? Great example. Podium is another great example.

Both of them, with the new products, have grown to $100 million-plus in a couple of years and basically just burned the bridges. This is legacy software. I think the more you fixate on, “How do we fix the business?” the less you're going to focus on, “How do we create a new business?”

So you've got to create a new business from scratch. You have customers. You almost have to say, “I'm going to be ruthless with migrating the current customers from the current business to the new product, even if it's a lower price.” It's the right thing to do, and you have to abandon the sunk-cost fallacy.

Help me on Podium. The $100 million in agent revenue triples to $300 million, and then it triples again to $900 million. If the price today is $5 billion that I'm paying, I'm paying for 2 years of triple-triple ahead of time for that asset. That's what it would be priced at in public markets.

For this business to work, we need the multiples in public markets to be way more than they are now. Do we not?

I think you need to assume that they will take over huge parts of the service budget in businesses and that they will not just be a billion-dollar company. I believe they'll be a multibillion-dollar company, because earlier, I think they were limited to one part of the stack and were on top of a bunch of systems.

Now they're taking over the entire software stack. That's the thing I like: You want founders who are ambitious enough to go after the entire stack, not just the earlier piece of the stack they were in. You want to be the only product that the company uses, and you want to replace as much of the digital labor as possible. That's the ambition.

So what you want to say is, “What's your market size here? How many people do all your customers have who are doing digital work? Do you have the ability to replace all of that payroll over time and all of the other things they're doing, and take a part of the payments transaction revenue?”

If you think that's a big enough opportunity, that's when you invest.

Do we see the total death of seat pricing, my friend? I hear you completely in terms of that movement into services. Does seat pricing die, and do we actually have consumption-based pricing as the primary pricing mechanism?

Seat pricing doesn't die. You know why? If you look at ChatGPT Enterprise, ChatGPT Enterprise is priced based on seats because seats provide predictability for enterprise buyers. But they don't drive expansion revenue by themselves.

So you basically have to bundle a lot more into each seat. ChatGPT, or OpenAI, sells seats based on different tiers where they have different functionality. I think Figma sells 3 different types of seats. So you're going to see different kinds of seats.

The big challenge is that seat-based pricing, which you alluded to, breaks when the product's core value is not about access but about doing the work on your behalf. At that point, charging per user doesn't make sense, because the user isn't the constraint anymore. It's the work output.

At that point, you have to go to outcome-based pricing. For example, if it's something like Harvey—I don't know how Harvey prices—I bet they price based purely on how many contracts they process versus how many people are using it. Even if 100 people are using it and they process 0 contracts, in theory, they should pay 0.

So I think you have 2 kinds of products: access products and work products. Access products are basically seat-based. ChatGPT Enterprise is a good example. Work products like Harvey are probably more outcome-based and not seat-based.

You mentioned Harvey there. We are seeing increasing competition within certain categories. If I think about law, it's Harvey and Legora. If I think about customer support, it's Sierra and Decagon, and there are dominant funded players.

How do you think about the ability for firms to king-make? Is kingmaking complete bullshit? Is it not? I'm intrigued to get your thoughts on that.

Kingmaking is a thing. I think it is a thing. You see earlier-stage companies getting these rounds that are valuing them at valuations that are really eye-opening.

That said, I think it won't work unless the company executes on the promise. Other firms can take it as a signal and decide to pile on or not. But ultimately, the company has to execute on the vision. If it doesn't, then it's just a bad bet.

So yes, it is there, but just because you king-make doesn't mean they are the king. They still have to execute and justify it. If not, then I think the reality is everyone's playing different games, right?

You and I know, being venture capitalists now, that somebody with a $10 billion fund is playing a fundamentally different game than somebody with a $400 million fund. If you try to play the same game they are, you're going to lose. You have to play the game that you're best equipped to play.

Benchmark plays a different game than, say, Andreessen Horowitz, but both of them play different games, and both of them do well at their game.

We mentioned Podium earlier and going to $100 million with that agent-first product. The growth is incredible, and the growth across this cohort of companies is—dude, we were doing this 8 years ago when you went from 1 to 10 at Slack, and it was like, “Holy shit, that's amazing.” Now it's at 1 to 70. It's still great, but there are quite a few who've done 1 to 10.

How does your mindset change around growth expectations for the companies that you invest in? Is a world of triple-triple, double-double dead?

It's not dead, but it no longer elicits the jaw-dropping awe that it used to a few years ago. As you said, 1 to 10 is becoming more and more common. Those numbers will get you—I mean, Lovable could probably go public with that kind of trajectory now.

The bigger question for me is durability, and it's not even quality—it's durability. Like you and I discussed, margins can improve over time and will improve over time. So it's not about margins; it's about whether the revenue is durable.

Retention is very important for me to understand. Are people using it? As I think we saw in the first wave, we saw many ChatGPT-like companies. There was a company called Jasper, not to pick on them, but they went from 1 to 40, and then they came back from 40 to 10 or something like that. Maybe 1 to 100 and 100 to 40 within a very quick timeframe.

There are a lot of tire kickers out there, especially in prosumer products, who test the product and then move on to something else. What you want to look under the hood, beyond all these numbers, are 2 things, which I think are the fundamental indicators of business quality: customer retention, or gross retention, and then net revenue retention.

Basically, those 2, I think, are the biggest indicators of quality. I would always take a company that's doing a crazy triple-triple, double-double with excellent growth and excellent net revenue retention over a company that's growing 10x in a year with really bad customer retention and less than 100%, or less than 90%, net revenue retention.

How do we think about ceilings on those markets? I’m specifically thinking about one that haunts me, which is Granola. I was one of the first ambassadors to meet Chris, and clearly Granola has crushed it and is an amazing product. Customer retention is sky-high, and revenue retention is sky-high. Honestly, if Anthropic or OpenAI did an enterprise product for note-taking that was connected to all of the different suite of products they have, I think that heavily threatens the market size Granola is able to expand into in large enterprise. How do you think about the worthiness of those retention numbers if there are alternative factors like that that could impact them?

Yeah, I think you want to basically weigh the retention in light of what they have encountered. You’re absolutely right. Just like you want to weigh the growth of a company in light of whether it has gone through any seismic events, if it hasn’t gone through any seismic events, you’ve got to take it with a grain of salt.

What competitive threats have you faced? Has a single competitor come out? Have you been able to ward them off? Has your retention stayed strong in light of that?

I do think some of these products, like Granola, are unique products that really open up non-consumption markets. We’ll see if that’s the case, but I would never have actually bought a note-taking product before—a separate note-taking product outside of Zoom or something—because Zoom comes with its own note-taker, and Gmail comes with its own note-taker. But Granola is so powerful that it basically got me to consume a separate note-taking product. It’s probably true for many of us, and I think that just changed the market opportunity for them.

I think Uber and so on are great examples of non-consumption markets.

Also, Gamma. You’ve got Google Slides built in.

Yeah, exactly. They’re a very good parallel. It’s a non-consumption market. You would never ask, “Why would you ever buy a PowerPoint or presentation product separately?” It’s a non-consumption market. It’s a zero market, but the product is so good, so remarkable, that it gets people to buy it separately as a separate SKU. Power to them. More power to them.

I think we need more of these standalone, remarkable products. Intuit is a great example, as you know. Microsoft tried to crush Intuit again and again and again back in the ’80s and ’90s by bundling everything into Office, but Intuit TurboTax survived and thrived.

Okay, but does that go against what you said earlier about the need to be multi-product? What you’ve done with Gamma is take slides out of Google’s G Suite and put them on steroids. It’s amazing, very deep, and feature-rich.

They can’t just be a single product and go. They will need to have a second product. I’m sure of that. They will need to have a second product.

What’s Gamma’s multi-product? I don’t know why Gamma wouldn’t create one. Just like they’ve taken their riff on PowerPoint, why can’t they have their own take on documents or slides in the same way?

I have to assume it’s basically the different kinds of content that people create, or even websites. It’s hard to say.

We’ll see. Both of them are in a wave of incredibly hot, attractive companies that have lower margins than we’re used to in traditional SaaS businesses. How has your mindset changed or stayed the same around margins? How should I think about margin assessment when looking at companies today?

Inference costs are dropping, so you automatically assume that margins, in theory, should go up. But I think it’s not about margins in years 1 or 2. The more defensibility or leverage you have over your customers and the choices they have, the more pricing leverage you have. I would rather see margins go up with price increases than cost decreases.

A good example is PayPal. Roelof Botha was on our board at Square, and he told us that PayPal raised prices 5 times in 3 years back in the day because there was such stickiness. They knew their customers really couldn’t do anything.

And you see—I don’t know whether Uber has raised prices or not, but I know that they’ve basically changed the economics of how much they pay drivers over time, so their margins have just expanded continuously. They’ve also raised prices in different ways.

I think there are 2 ways of increasing margins. First of all, you and I both don’t look at margins in years 1 and 2. It doesn’t make sense, even in years 4 and 5. On one side, you want the ability to increase prices. On the second side, you want the ability for costs to get lower. I think that second thing is happening by nature.

What you want to see, in addition, is the ability to have such a good product—and ideally a multi-product story—that makes switching really hard, so you can raise prices.

Can I be blunt? In the environment that we’re in today, the 2 things that you said there—the ability to increase prices and margins in years 3, 4, and 5—given how much the world is changing, I have no idea about their margin structure in years 3, 4, and 5.

Yeah. It’d be premature to look at margins. You don’t look at margins; you look to see whether or not they have a product that is compelling enough. Obviously, Disney+—I think every year they increase prices on me. They’ve gone from $7 to $10 to $14 or something like that. Amazon Prime is another one.

But this takes many years. You want the potential. You want to evaluate the potential. Do they have the potential? Do they have the ability to increase prices in the future?

It’s not something I worry about. I think durability and defensibility are much more of a worry. I think good companies, if they are defensible, will have the ability to increase margins. That’s what I was saying, actually.

Have you ever shopped in Chanel Coco?

I’ve gone into—I think they do have stores, many stores. I’ve never shopped myself. My wife has.

I buy my mother a Chanel handbag every year for Christmas and her birthday. Do you know what they do every 6 months? Prices go up 10% every 6 months.

10%? Dude, 10% for the same product?

I used to buy a handbag and it was £500. Now it’s £10,000. Holy cow. My question is: how should we invest in LVMH, clearly?

Durable, defensible, 100-plus years, right?

100%. And, dude, in a world of increasing wealth inequality—actually, that’s an awful statement—can you have good businesses selling to non-wealthy people? You’ve worked in fintech before.

Robinhood is a good example. I think you’ve got to have massive scale. I mean, I’m in an investment company called Atlas, which sells to billionaires. It’s the opposite. It’s a great business; it’s just that you need massive scale.

You need a wedge product that is almost cheap or free. Look at Robinhood. You have to have free, canonically, in your thing. Robinhood already offered free stock trading for a long time, and that was their core pitch. That allowed them to get a lot of people.

You need to have something free or some hook that is really low-cost, which allows you to expand. But it is a harder business because your ARPU, or average revenue per user, is low enough that you need millions, if not tens of millions or hundreds of millions, of people.

When you sell to very rich people, wealthy people, or large enterprises, look at Palantir, which is the business equivalent of selling to wealthy people. They have, I think, fewer than 1,000 customers—maybe even fewer—and each customer pays them $20 million, $30 million, $100 million, or $1 billion. It’s an easier business. Obviously, it’s a very hard business, but you can say, “I like those businesses.”

Look at Veeva. They said they went public with 400 customers. Four hundred customers.

Do you bother to do market sizing today, given the transience of markets? As you said, some of the best companies make you pay for things you never thought you’d pay for. Do you bother to do market sizing?

I think non-consumption is the biggest challenge, but yes, you can’t not do market sizing. I do bottom-up market sizing for the specific segment, and I always know that any customer base with more than 10,000 customers—or even a few thousand customers—you’ve got to segment them.

There’ll be a few different segments, so you want to understand, within each segment, what the bottom-up propensity to pay is and what problem you’re trying to solve for them. Then you’ve got to talk to them to understand what the budget is. You’ve got to do the work.

What’s your biggest misread on market size, and how did it shape your thinking?

Shopify. I remember seeing Shopify at $1 billion, and I was like, “How many e-commerce merchants are there really?” Or maybe it was even before $1 billion, in one of the early rounds. I felt really concerned.

What I missed was that Shopify was not just selling e-commerce. It was basically allowing anybody to sell. It changed the behavior of any entrepreneur on the planet. Anybody who wanted to sell something went to Shopify.

It wasn’t just existing e-commerce merchants, and that’s what you want platforms to do. They literally make it possible for every person to think about selling something, renting their home out, or taking a ride—things they never would have thought of before—or installing or buying a new presentation app or a note-taking app.

They are the biggest hits, but they’re also the biggest misses. If the bet doesn’t play out, they’re screwed. If the bet plays out, they could be bigger than anything else. Google is a non-consumption example. Many of them are non-consumption markets.

They're new behaviors that didn't exist before. That's, in some ways, what venture is all about. It's not about what's existing; it's about new behaviors and betting on them. Facebook was a non-consumption market, right? Think of all these iconic companies. The thing that's amazing with Facebook is how easy it was for you to dismiss it as the 52nd social network.

There had been Friendster, MySpace, and so many others before it. The defense was identity. On Friendster and MySpace, you didn't know who the people were. They didn't have real photos. I think I remember when Facebook tried to go into Japan, Japanese cultural norms meant that all the Japanese social networks back then were incognito. You couldn't, for some reason—maybe it was about saving face or something—share your real name or your photo. So everyone said, "Facebook, you've got to adhere to Japanese cultural norms. You've got to change Facebook and make it similar." Mark said, "Absolutely not. It may take us longer," and they succeeded.

How do you prevent prior wins or losses from impacting future decision-making? My biggest mistake is that I lose all my money in a market, and it inherently makes me attracted or not attracted to it in a way that could subvert decision-making. How do you avoid that?

It's very hard. I think it's a mental thing where you have to take every opportunity from first principles. We all struggle with it. I think the best venture capital—I was asked what the best venture capital bets are, and I talked about a paradoxical one. I think it's Mike Moritz betting on Instacart.

Why? Because he lost $370 million on Webvan less than a decade ago. He burned it, too. Same space. Apoorva comes to him, and he bets on it. He bets on it after losing hundreds of millions of dollars, with the whole thing having burned to the ground. Think about the first-principles thinking and the courage needed to make that bet. I think it's brilliant.

You've got to laugh at being a Sequoia partner. You're thinking, "You're not going to bet on food delivery again. Come on, Mike. Really?" I'm so curious to see how he reacted. You're like, "We know you did Google and we love you, but come on, not food delivery again." Main Street shopping, exactly.

The market is one way we trip ourselves up: "The market's too small, the market's too small." The other one that I always make mistakes on is price. How do you think about this when you reflect on having done so many good deals? In your experience, are the best deals the most expensive?

I think there are 2 ways. I've now realized, after many years of doing this at seed and Series A, that price almost doesn't matter if you're right about the company. I think you just invest at whatever the price is. For example, I invested in the seed round of Faire about 8 or 9 years ago at a $20 million valuation. It was a very expensive seed round at that time. It was the highest-priced seed deal at that point. I think it's been 100x or 200x for me.

I think you invest in a company when it's great and you have conviction—you invest. Now, I think at Series B—at B-plus—that's when price starts destroying returns. By then, you've got real revenue and real traction. You can pick a generally good company and still get crushed. For example, one of my friends invested in this security company when it had $100 million in revenue. He invested in it at $4 billion. I think they've gotten to $500 million in revenue, but guess what? They're still at $4 billion. So, basically, they will not make 1x the capital they invested. That's the challenge, I think.

But guess what? Even in WeWork, Benchmark made money. Benchmark made money at WeWork because they invested early enough. So, I think at sub-$100 million—and maybe that's an arbitrary number—if the company is good, you'll make money regardless.

You mentioned WeWork, which we'll get to, seeing as I used it as an example in a show we did with Miles Clements from Excel. I just want to touch on the Series A market and what you're saying about pricing and why it matters or doesn't. I'm with you 100%, but we're seeing 100x ARR for $3 million-revenue companies, and they're being priced at $300–$400 million in this new environment.

How do you advise me as a Series A lead investor to operate in a market where Series As are no longer $10–$20 million at $100–$150 million valuations? They're actually $300–$500 million, with $30–$50 million rounds?

I don't think a Series A investor can do deals at $300 million or $400 million. There are 2 kinds of deals that investors have to do. One is where there is less legibility on the company, where you're betting that there's some early product-market fit. There's not $3 million in revenue; there's half a million in revenue. That's, like you said, when you get it basically for $50 million or sub-$100 million at $1 million or so in revenue.

But then, as soon as you get to $3 million or $4 million in revenue, it gets tough. Maybe you can do a couple of deals like that, but I don't think you can build a Series A fund doing deals at $300 million or $400 million because there isn't going to be enough ownership. Some of these are going to fail, et cetera. Most deals, I think, you've got to have double-digit ownership. You've got to invest slightly earlier. It's a tough game, but you've got to be patient.

The good news is, like I said, concentration can be your friend in some ways. Then you don't feel that you've got to do 10 deals a year. You can do 4 deals a year and have 15 companies in the portfolio.

Dude, you've got $250 million in the fund. You've got $200 million when you actually look at investable cash. If you don't have any reserves, you've got, say, $10–$20 million Series A checks. If you're wanting to get double-digit ownership, which we all say that we want, is that enough? I'm not being cynical; I'm asking for my own advice. Is that enough?

We have 35% reserves, so you have to have a mix of seed and incubation bets and a mix of Series A bets. The incubation bets are bets you take on founders who are basically the best in the world at what they've done. A good example, I think, was investing in a company called Lead Bank. This was before Marathon, but with my Marathon partners. They led the round with Vinod Khosla and Mickey Malka at Ribbit in a company that was my colleague Jackie Reses's.

Yeah, I interviewed her. She's amazing.

That was its inception round at some crazy valuation—I mean, a very strong valuation. So, not a crazy valuation, but a strong valuation, because Jackie was Jackie, and she had built the bank at Square. She built this bank again.

You want somebody in an industry where they know the inner workings better than anybody else in the world, and you say you back those people. That's a much, much better risk-reward there. So, you want to do a few of those in addition to a few incubation and seed bets, in addition to Series As.

And you're doing that, right, in some ways. I think you've got to mix it up. As an early-stage fund, I think every firm—I think every early-stage firm—you've got to have access to founders and be their first call when they go to start a company. Then you meet founders you don't know beforehand, kind of betting on traction and so on. You've got to have a mix of both kinds.

Do you buy the proprietary founder access? Again, this is where I get grumpy, but I've done 3,000 shows, dude. At some point, you have to get cranky. Every venture investor sells the proprietary—

Proprietary founder access. What is proprietary is your ability to add value. I think founders—you basically have to build. If you're just a source of capital and assuming founders will come to you, you're not going to win. What you have to offer them is something very different and unique.

We all have to hone, as investors, what it is that we're offering. Is it counsel? Yes, that's free. Is it distribution? You offer incredible distribution. Is it a network of customers that you can get them access to? Is it talent hiring? What is it they can do with you?

Do the best founders need you? Keith Rabois always says the best founders do not need a venture investor's help. You've worked with the best. How do you feel?

They would not need it. I don't know. I generally agree with Keith that, on the margins, investors don't add value, and the value they add gets less and less as a company grows. But I do think there are a few points where there are a few things you can do on the margin. For example, helping them choose between this candidate or that when they're hiring, or helping them think about go-to-market. That could make the difference between the company being a mediocre exit or outcome and being a generational company. You don't need to do everything for them, but just those 1 or 2 things that you can help with on the margins can hopefully be the difference.

What would be your advice to LPs when they are consistently sold by GPs like me and you on proprietary founder access? "I have the best network. I'm a super-smart fintech expert, so I know it better than anyone." What would you advise them on manager selection when everyone says proprietary founder access?

Very simple: go and talk to the founders. Go and talk to the founders and see why they chose them, especially the earliest-stage founders. Go to the seed founders that they invested in at the inception or seed fund and ask them, "What is different? Why did you choose them? What other options did you have?"

You've got to use the data. You can't just rely on everybody.

You're right. Most VC pitches look the same. What you want to do is dig one level deeper and talk to the founders themselves, understanding, for each of the last 5 companies they've invested in, why this founder picked this firm.

Can I push back on the model that you have and just pretend that we're a hypothetical partner, okay? You have 35% reserves. Why is that optimal over just having more lines in the portfolio? When you hear about a 100x or 200x multiple on a fund, I'm like, focus on ownership, have more, increase diversification, and take away the reserves. Why do you think that's too risky for operating?

I think that's too risky for operating. I'll give you an example. The Trade Desk, where I'm on the board, had 2 seed investors: IA Ventures, Roger Ehrenberg, who's absolutely a GOAT, and Founder Collective—again, goat farm. Both of them had completely different philosophies.

Founder Collective only does first checks. They never do any pro rata afterward, period. Roger, on the other hand, doubles down again and again and again. The Trade Desk raised, I think, 2 or 3 rounds of financing. That's it, and it went public very early. It was very hard for them to raise financing.

The multiple that Founder Collective generated was incredible because they only invested at the seed round, and they got it at $5 billion or something like that, or even more. I think they held for longer. Roger generated a huge dollar return, even though his multiple was different. So there are 2 philosophies.

I do think that if you look at my philosophy, it's more like—if you look at Founders Fund, which I think is one of the best-performing funds, a huge part of their success is doubling down on the companies that matter. I think the unsung hero of Founders Fund is a guy called Napoleon Ta, who leads the growth practice. Napoleon is basically the one who decides which companies they should double down on.

If you double down properly, it changes the composition of the fund because you have much more insight. I would argue that if you work closely with these founders, you have much more insight into these companies and how they're going to do, and even how they think about the future opportunity, because you've thought about it with them, than you do with a random company you meet.

Now, there is a balance there. You don't want to be overconcentrated, but I would argue that if you have X number of bets and you work with the founder and think highly of them, that's important. That's why each Founders Fund fund is named after the company that makes it, in colloquial terms. There's the Anduril fund. There's a SpaceX fund, et cetera. Why? Because that one company is the one that makes it.

That's a reality, Harry. Literally, you have 1 company—most likely 1 or 2 companies—that will drive most of the returns of any given fund. The question is, do you just want to have the initial stake? Do you want to increase your probability of finding the initial company? Or do you feel like there's an explore-exploit thing, right? Where do you stop exploring, and when do you stop exploiting?

Different people have different points of view there. My first fund could have been the one you're now backing, and I'm very grateful to you for supporting me—very much, 18 or 19. But it could have been, at one point, the Hopin fund. It could have been the Clubhouse fund. It turns out that it will most likely be the Linear fund. I think it will be.

I think Linear is a great business, and we are very early there. But my point, with the transitions in name, is that it wasn't obvious. My question to you is: with preemptive rounds coming so fast, how accurate do you think you can be in predicting the winners? Because it definitely wasn't obvious to me.

You've got to be thesis-driven first and foremost. What we are is—we think about what the thesis is. In other words, you've got to have a good sense of who the other companies and players are in the space, and you've got to understand why this company is better than every other company. On what dimensions is it better, and is that dimension durable enough over a venture time frame, which is 7 to 10 years or even, maybe, 10 to 12 years now?

You've got to do the work. You've got to be thoughtful and patient. Remember what being concentrated does? It gives you more time. It gives you more time to meet companies. It gives you more time to think. It gives you more time to be helpful to companies. But you don't feel the pressure to deploy on a monthly basis.

If you look at a 30-company portfolio fund over 3 years, which is the initial deployment period, you're basically investing in almost 1 company a month. I almost feel there's pressure on the folks who do 30 to 40 to do basically 1 company as a partnership per month.

If you look at Greenoaks, their 6 or 7 funds have basically, what, 65 companies overall—around 11 companies per fund. He's an absolute beast. He's got 10 companies that have returned over $2 billion. The incredible thing about my life, Gokul, is that I hang out with these people and I just leave feeling like a total loser. You leave Micky Malka and you're just like, “Yep, no, I didn't do the Robinhood round.”

Yeah, you've just gotten better with every one of those interviews.

I've seen your style and your investing. Who do you learn from?

Entrepreneurs—most of the people in the arena. Since I'm so—and our firm is so—trend-driven, and we really care about what the thesis is and what the market is, the best way is to talk to people who are in the trenches building products.

You can talk to investors, but they're always 1 click away. You've got to talk to people who are building products. You've got to understand what's changing in their lives and how they're thinking about the customer. In today's world, you've got to stay close to the model companies, for example.

I have people that I meet with at each of the model companies to understand what's coming down the pipe, how they're thinking about the world, et cetera. You've also got to understand—I always used to joke, and this was 10 years ago—if you're a startup that's directly in Google's roadmap, you should not be building it.

Google is very good when something is directly in its roadmap. They're like a tank. They'll just roll over you, even slowly. It doesn't matter. They were implacable; they would roll. But if you are even 10 degrees to the side, it's very hard for them to move like that. You're generally safe because they're just rolling in 1 direction.

You want to know what direction the turrets or the guns are pointed in for each of these models. One part of this is what the labs are going to do. The second part is what customers are going to do and what customer behaviors there are.

You want to talk to 3 or 4 companies in the same space. I'll never forget that people who met Tony at DoorDash, and many of the other folks who were fundraising at the same time, always felt Tony had a deeper understanding of the same market than others globally. You basically want to meet them.

That's why one of my biggest regrets is actually passing on Vanta. I met Christina, but I had already committed. I was like, “This person is going to win the market,” but I had already committed to another company in the space.

Sadly enough, once you invest in a company as an angel, I didn't have the time to scan the landscape and meet with all the companies in the space. That's what I do now. I have time, but Vanta will definitely be part of my anti-portfolio. I think it's part of yours, too, if I remember correctly.

Elad sent me something and he was like, “Dude, this is amazing. This is amazing.” Every time Elad sent me something, I said, “It's amazing. Just fucking do it.” “Do not think you're smarter” is my takeaway there.

There are other people who do it and think they're smarter. I think I'll leave them out of it, but there are some friends where I'm like, “You consistently sent me this.”

Anyway, we mentioned WeWork earlier. I use WeWork as an example with Miles Clements from Excel about selling. I'd love to hear your thoughts on how you think about when to sell. Obviously, we have investors, you have LPs, and they care about DPI today more than ever. How do you think about liquidity, and when's the right time to take chips off the table?

As an angel, I used to hold until IPO. Figma had many liquidity opportunities during the years, but I kept holding it for 13 years until it went public. I think at the IPO it was actually priced very nicely. Unfortunately, after the IPO, it has been more challenging price-wise.

As a fund investor, it becomes interesting. I think there are 2 situations. One of the things I think most early-stage firms get wrong is that they just focus on MOIC; they don't focus on IRR. MOIC is multiple of invested capital. I think IRR matters a lot, as you know.

An LP once told us about a firm that gave them a 7x MOIC over 20 years, and that was a teens IRR. That's like, okay, there's something crazy here. I mean, that is a venture, venture, venture firm. You've got to look at go-forward IRR and your projection.

If your go-forward IRR at every liquidity opportunity is lower than what you're basically promising your LPs, or what you think your fund should have, I think you should sell.

I think you have an obligation to LPs to at least sell. I like Fred Wilson's strategy around selling: sell a third, hold a third, and trade a third. This is when the asset is completely liquid. But in this case, I think you want to sell at least part of it, especially if the asset is an equity—a company that will return a chunk of your fund.

So, if it's going to return 20%, 30%, or 40% of your fund, I think you owe it to your LPs to sell a piece of it, especially if the go-forward IRR is not compelling. On top of that, you then have the whole period after the IPO where you have, you know, obviously your stock

Exactly, and you don't have [__]. Exactly, and that's another uncertainty. So I do think the secondary markets have been one one best almost interesting developments over the last few years.

Now, all these great companies have pretty liquid secondary markets where you can sell shares. Obviously, there are ROFRs and so on that the company has. I do think there are these hyperliquid periods in the market, and now is one of them. So, I do think one should be very careful and thoughtful about what the go-forward IRR is for any asset one owns and really think carefully about whether one should sell or not.

Can I ask you, when you look back at the angel portfolio, what's the biggest regret? Not investing more?

I think a good example is Quince, which recently raised at a $10 billion valuation. I saw Quince 4 years ago when it had a $100 million valuation, and I thought, "It's a D2C company. D2C companies were kind of on the downswing. How good can this company be?" So, I literally just dismissed it. I didn't even look deeper into the company.

So, what was the takeaway from that, then?

The takeaway is that you can't just take an industry and say it's good or bad. Within every industry, within every category, there are great companies and there are mediocre companies, and you've got to understand each company's remarkability and differentiation.

Quince, for example, had an incredible 35% to 40% repeat purchase rate, which was higher retention than most consumer apps. I should have paid more attention to that versus dismissing it. It was literally in the blurb, and I was like, "Okay, so what? How good can this get?"

That's non-consumption. What I don't like, respectfully, about what you said about your trend style of investing is that some of the biggest misses I also have are when there's an amazing founder who's clearly amazing but operating in a bad space, and they pivot 3 months later into a good space, but I turned them down when they were in a bad space.

I'm like, "I don't care what they're building. I don't care what trend it is. Gokul, you're amazing. If you're selling pillows, I'm in."

If you're a seed investor—if you're a pure seed investor—I think you have to do that. I think that's what Y Combinator does, right? I think that's actually the right way to do pure seed investing.

First Round Capital, I think, is one of the best seed funds. Guess how many companies they have in each fund? 80 companies. Why? Because you've got to take 80 bets. Y Combinator, of course, is the best seed investor of all time, or pre-seed. I mean, you've got to take hundreds of bets because most companies will pivot.

But I think as a concentrated portfolio, you've got to understand the business. You've got to bet on both the business and the founder, unless, again, the founder is an N-of-1 founder in a space.

I think there are 2 categories of founders. Actually, there are 3 categories of founders in some way. There are repeat founders who know a space exceptionally well and have done it before. You're betting on them again and again to do it.

Cybersecurity is a good example. Cybersecurity is full of repeat founders. They know the market, they know the history, they know the customers—all of that. But then there are consumer companies. Consumer internet is full of first-time, extraordinary founders. Mark Zuckerberg, Larry Page—all of these are just first-time founders.

So, those are 2 archetypes. The third archetype that has come up is AI lab researchers. I think that one is a more interesting bag, where you have, of course, Dario Amodei and the OpenAI folks, but then you have a bunch of other labs that came up and ultimately didn't turn out to be anything.

For these kinds of folks, as a seed investor, you just want to blindly write a check into them: a brilliant young person who's done something extraordinary, a repeat founder, or maybe an AI lab researcher. You just want to bet—

Would you invest in a frontier model or AI lab—labs where they're clearly fucking amazing people, pedigreed to the hills, but the price is in the billions?

Not possible. I don't think with our fund it's possible. I think the ownership is just literally—the first round for these companies is, like you said, $1 billion. So, I think it's just too high. The risk-reward is just not worth it.

Are you seeing the mega funds cannibalize the business model of our Series A?

They're playing a different game. Look, I think very highly of the mega funds. They've basically gone and deployed $15 million checks almost as an option and as lead generation for the next round. Their strategy is obviously to have an index at the Series A of every single good Series A company, and then double down on the ones that truly matter, triple down and quadruple down, do SPVs in them, do specialized funds in them, and so on.

I think it works. It works for LPs. In some ways, it's a different asset class than early-stage funds, but it's different. I think smart founders, good founders, have started to look beyond just the fact that they can get $10 million very quickly from a mega fund and say, "What am I getting here?"

We see many examples in mega funds of partners leaving the fund and the companies being orphaned within the mega fund because their partner has left. Now they don't have a single person to advocate for them in any way, shape, or form, and they're adrift.

Anybody who's a repeat founder is most likely to see behind just the glitz of a mega fund in some ways, because many of them have gone through it, especially the ones that have started a company in the last 5 or 6 years. We have many stories where the mid-level partner at a mega fund has left, and they're like, "Okay, I don't have an advocate in the fund, and now I'm left with a person who I don't know, and they don't know me, and they're joining my board. That's a tough one."

You've seen Nico Bonatsos, Matt Cozzol, and Arif Khan Mohammad. We've just seen the start of this continuing wave, and there will be a huge amount more spinouts. Or do you think we're going to see that curtail?

I think we're going to see more spinouts. I do think there is a limit, but I do think you're going to see these mega funds train yet more waves of investors, and these investors are going to realize that being a mid-level partner at a mega fund is not what it's cracked up to be.

So, they're going to spin out and go back to the way of doing things in venture that used to be the way 20 or 30 years ago, which was a small group of partners building deep relationships with entrepreneurs.

Listen, dude, I want to do a quick-fire round. What have you changed your mind on in the last 12 months?

I used to think pure remote would scale for early-stage companies if you had the right culture, but I don't think that anymore. I think you've got to be in person at least a few days a week.

What made you change your mind there?

I think it was just seeing a few companies where the companies literally died because the founders were unable to agree. They had everything going, but the founders were just not in the same place and were not able to move fast enough, agree, align on the strategy, and change things.

So, iteration speed suffers massively if you're fully remote. It doesn't need to be 5 days a week, but at least 3 days a week.

Biggest advice to a young person leaving university today?

I know it feels exciting to start a company. Everyone's doing a startup, and AI is the new wave, but my strong advice is to first get 2 to 3 years of work experience at a good company. You won't regret it. You'll learn a lot.

Both the experience and the network of people will be invaluable for you. So, just 2 or 3 years. Don't be impatient. Life is long. Get some work experience before starting a company.

You've got to answer an unfair one. You've got to invest in 3 different types of funds: a seed fund, a Series A fund, and a growth fund. Which 3 are you choosing? And they can't be mine.

First Round Capital and Benchmark. I've been an LP in both, and then Greylock. First Round would be the seed, Benchmark would be the Series A, and Greylock would be the growth.

Well, that wasn't a hard one, was it? You thought about that. Most people are like, "Oh, I can't do that."

A lot of LPs have asked us this question. I've actually answered this question for LPs before.

Dude, that's fantastic. What has been the hardest decision that you've made in your career? You've left amazing companies. What's been the hardest decision?

Leaving Google. Leaving Google—I think there was a saying: "You leave Google only once." So, leaving Google to start a company. I was on a pretty incredible trajectory there.

I was learning a lot and really enjoying it. It was really tough to leave Google. I don't regret it, but it was very, very hard to leave Google.

Who is the best CEO you've ever worked with?

All 4 of them: Larry, Mark, Jack, and Tony. And now Brian Armstrong, Bill, and Ben at Pinterest. It's a hard one. I think all 4 of them have different superpowers.

I would say the best technical CEO is Larry Page. The best growth-centric CEO is Mark Zuckerberg. The best design-centric CEO is Jack Dorsey. And the best physical-world operational CEO, and the one most likely to be the next Jeff Bezos, is Tony Xu.

I'll take that. That's a good one. What's the biggest miss we've said Braintree? Is it Braintree?

Quince, man. Most recently, Quince, but to be honest, an even bigger miss than that in some ways is that I couldn't predict that Facebook could be a $2 trillion company.

When our company was going to be acquired by Facebook, I was arguing with the corporate development team at Facebook, and we were arguing over the terminal value of Facebook. We had to put China into the mix, saying, "This company in China will get us to $40 billion in market cap." We were arguing whether it was $20 billion or $40 billion several years from then. This was in 2010.

It turns out that in 11 years, it was a trillion-dollar company. When these things work, they work at a scale that is unimaginable. Even at Google, I remember very well after the IPO—I was there during the IPO—and we were sitting around with a bunch of PMs, saying, "Man, the company's valued at $30 billion. It's too expensive. Too expensive."

These things just compound, and it's incredible to see them become trillion-dollar companies. So, Facebook and Google, in some ways, are the biggest misses in terms of not being able to predict that they were going to be multitrillion-dollar companies.

Which angel investment is the highest multiple?

Fitbit.

What was the multiple?

I think it was between 500 and 1,000× at the time of the IPO, but it has sadly gone down since then.

500 to 1,000×. Jesus Christ. Tell me, final one: what most excites you about the next 10 years?

I like to be optimistic. I think we have too much pessimism.

What do you like? What's the thing where you're really freaking pumped about it?

The most ambitious entrepreneurs are finally tackling the hardest problems. I feel the ambition that AI, especially, has unlocked is just incredible. The ambition of entrepreneurs tackling the hardest problems facing humanity and society is absolutely incredible.

How can you not be optimistic when you have Elon going? I think we now have entrepreneurs who are role models, who are not just building small companies, but are truly taking on humanity's problems.

The answer to Peter Thiel's question of, "We were looking for flying cars, and we got 140-character apps," is finally coming into focus.

I've got to ask one more. You mentioned Peter Thiel there. Obviously, he has the Thiel Fellowship and a preference for young, ambitious founders. We've seen this massive movement toward very, very young founders. We mentioned McCauley who are brilliant. Are you in line with the shift toward the earliest, youngest founders? How do you feel about that shift to super-young founders?

I actually am a huge fan of it. Even at companies, I feel some of the companies that are not hiring young people are making a huge mistake, because young people are more AI-maxed—as you could call it, like looksmaxxing, AI-maxing—than anybody else.

In fact, I think younger people are adopting tools better, and they just live and breathe differently than others. I am a huge fan. I've actually invested in more dropouts as an angel over the last few months than I have invested in during the rest of the last 15 years I've been investing.

I don't think it's the right thing, to be honest, for many of them to be dropping out and starting. I do think they could benefit socially and emotionally, et cetera. But some of them are just exceptional. I don't think all of them are, but I do think this crop is going to produce some incredible founders.

Gokul Rajaram, I appreciate you. I've got so many notes, I had to go on different sides. This has been fantastic, so thank you so much for being so amazing, dude.

My pleasure, my friend. I look forward to doing stuff together.