a16z, Anish Acharya: Is SaaS Dead? Do Margins Still Matter? Why We Are Not in an AI Bubble?

You have this innovation bazooka with these models. Why would you point it at rebuilding payroll, ERP, or CRM?

The general story that we're going to vibe-code everything is flat wrong, and the whole market is oversold on software.

Anish, dude, I've wanted to do this for a while. We've been going back and forth, and I'm so glad that we can do this in person. Thank you for joining me.

1. Why building an AI company requires being in San Francisco

Of course. Thank you for having me.

I'm diving right in. We were just chatting, and I was saying I think it's better to build in London than in San Francisco. In places other than San Francisco, talent is cheaper, it retains for longer, and you don't have the promiscuity of people jumping from role to role. You've built a company now both in Canada and in San Francisco. How do you reflect on what I just said?

I disagree with you. I wish it was true. I simply wish it was true, and I want it to be true, and maybe it will be true. We always love to say that talent is equally distributed and opportunity is not. The truth is that cities are the original network effect, and for technology, there is a network effect for builders in San Francisco.

For this moment in technology, where so many of the secrets are these things whispered down shadowy hallways, the benefit of being in San Francisco is enormous. There's also—we just talked about this—a selection-bias question: do you care enough to make it happen in San Francisco? You can make it happen anywhere: New York, London, Toronto, Tel Aviv, you name it. But there's something different about saying, "I'm going to give everything else up, be singular in my focus, and move everything to San Francisco to make it happen."

Are there any other locations where you think there is actually positivity associated with being located there?

Tel Aviv. I think in Tel Aviv, you can be incredibly ambitious and uncompromising on that ambition and have a really, really good reason to be there. I think the other nice thing about the Tel Aviv ecosystem is that the country is so small—it's 10 million people—that you can't possibly fool yourself into thinking that the domestic market is going to be big enough for whatever you're doing. So you immediately go outside.

Whereas, if you're in the UK, there are 60 million people here. You might say, "Well, that's actually a lot of people." And you know what? There are parts of the market, like fintech, where the LTVs are so high that perhaps 60 million is sufficient.

But for most mass-market products, it's just not sufficient. If you end up starting focused on the domestic market, it's often hard to actually move on to a bigger market. There are incredible counterexamples, like ElevenLabs, but I do think it's just that much easier in San Francisco, and that's why that's where I focus.

You said the word "sufficient" there. Yes, you can build a sufficient-size business, say, in the UK—a $3 to $5 billion business, for example. Respectfully, when we look at companies being created today, $3 to $5 billion just doesn't seem like it's interesting enough. Has the world of venture changed so significantly in terms of what is sufficient for a venture outcome?

$3 to $5 billion is an extraordinary outcome, don't get me wrong. In no way am I minimizing that. And, look, I do think that those types of venture outcomes stack to create really meaningful funds. So this is not about working backward from venture economics.

2. The "SaaS Apocalypse" myth: Why "vibe coding" everything is a lie

But the biggest companies in the world today are trillion-dollar companies. If you want to build a trillion-dollar company, if that's your intention, you've sort of got to start with a set of assumptions that can lead to that. If your intention is to build an extraordinary enterprise and you build a $3 to $5 billion enterprise, you are one of the few people in the world.

When we say those $3 to $5 billion companies—

Yes. I heard a brilliant statement, which is the SaaSacre—the massacre of SaaS companies—that's going on in the public markets today. Bloomberg is trying to get "SaaS apocalypse" to stick. When we look at it, essentially, investors are no longer confident that traditional enterprise revenue is sticky or durable.

Are they right to question whether traditional enterprise revenue is sticky and durable?

I think software is completely oversold. I think it's a silly story. Look, if you look at SaaS spend today, if you look at IT spend overall, it's 8% to 12% of enterprise spend. So even if you vibe-coded your ERP and your payroll, with all the risks and dangers that that entails, you're going to save 8% to 12%.

You have this innovation bazooka with these models. Why would you point it at rebuilding payroll, ERP, or CRM? You're going to take it and use it to extend your core advantage as a business, or you're going to take it to optimize the other 90% that you're not spending on software today. I just think that, of course, there will be secular losers. There are specific business models that are now going to be disadvantaged, but I think the general story that we're going to vibe-code everything is flat wrong, and the whole market is oversold on software.

3. How AI agents are finally breaking the lock-in of legacy software providers

Okay. So we are actually overly negative, and we're being too critical on these companies. How do we think about the continuing negative growth that we've seen in a lot of these companies, and the continuing seat contractions in a lot of your CRM providers or Monday.com?

I don't know if that's what we're seeing across the board. I looked at the data this morning, and if you look at public-market SaaS companies, 75% have raised prices since ChatGPT was released. Seventy-five percent. They've raised prices meaningfully. The mean is 8% to 12%, but there's a large group that have raised prices by 25% or more.

Is that not because they have to? They're not growing seat count, so they have to grow revenue.

Price is a measure of product-market fit, right? If you have enormous competitive pressure, you're not raising prices; you're typically cutting prices. So I think, one, you've got this sort of dissonant fact that prices are going up.

Two, if you look at the incumbents today, ServiceNow is not IBM. They're a highly capable incumbent. They just went public, and they raised guidance. I think it's very easy to look at these things and say, "Incumbents, incumbents, incumbents." Again, they're not Sears; they're very, very capable, and I think they actually have a right to win and deploy technology in the context of these workflows.

Now, will there be disruption? Of course. We've talked a bunch about companies that were once priced on seats, which are now going to be priced on outcomes, and that is going to be a big drag. But I think for the majority of SaaS, it has so little upside in being rewritten and vibe-coded, and so much downside. Why would you do it?

An interesting topic that's not discussed is the cost of transitioning from one SaaS provider to another going dramatically down. So, systems integration.

If you have an SAP system, you are a hostage of SAP, and they need to do nothing after they win you as a customer except the bare minimum. If you want to switch to Oracle, oh my God, that's a multiyear, high-risk process. It's probably going to fail, and you're probably going to get fired. It doesn't happen.

But now, with coding agents, the complexity of transitioning from SAP to Oracle is dramatically lower—the speed, the risk. So that is how I think coding agents show up in enterprise software, especially among public names: decreased switching costs, more customers, and fewer hostages, which is a positive incentive for the entire ecosystem.

4. Incumbents vs. Startups: Who actually wins the AI distribution war?

You mentioned Alex Rampell. It actually sounds super weird, but I actually think about Alex every day. Yeah, well, there we go. He says the most brilliant thing, and I'm going to butcher it slightly: "Will the incumbent acquire innovation before the startup acquires distribution?"

That's right. Yeah.

How do you think about who wins in this world? Is it the public SaaS company that has distribution, be it HubSpot or Salesforce, with millions of customers? Or is it actually the startup that has speed, agility, and incredible engineers?

If history is any guide—and, to reference Alex, he would say that it often is—those who have actually studied history tend to do better than those who have not. When you have this product cycle and a capable incumbent, what happens is they usually make their product better for their existing categories.

Microsoft will make a better word processor than they've ever made. Google will make a better search engine than they've ever made. We're actually starting to see some of that, anyway.

What you instead see is the native categories that did not exist before the product cycle being owned by startups. So I think that's a little bit of what we're going to see.

You know, if you said something like software for movies, AI movie making, or sort of AI-assisted movies, that's just not a category in which there is an incumbent. I'm betting that a native company will actually win that. It probably won't be Adobe. Will Adobe make a better Photoshop and Illustrator than ever before? Probably.

Right. You said there about native being an opportunity in terms of where opportunity sits in the stack. Why do you think the application layer will create more value than foundation models?

I don't know if it'll create more value, but I think it's underdiscussed how much value it's going to create. If we lived in a world where we had a single foundation model company—which, at the time, was OpenAI, which was a whole generation ahead—then they essentially were this unique supplier to everybody downstream in the innovation ecosystem.

They could do what you would do if you, for example, controlled the Beatles and were the only record label that had the Beatles. It's like, do you want the Beatles or not? You can charge 99% of your customers' gross margin, and you do. You actually tend to charge 100% or 110%. So that was a big risk to the ecosystem.

What has instead happened is that we have all these foundation model providers. They're all innovating roughly in lockstep. Eighty percent of what they do, I think, are actually substitutes for one another, and then there are the open-source models, which also do the same things.

In the 20% which, arguably, is where a lot of the value is, they are all specialists. Because you live in this world of multimodel, where for some use cases they're substitutes and for some use cases they're actually specialists, there's a lot of value in having an aggregation layer, and that is the apps company.

Let me tell you about 2 categories specifically. One is coding. I think that if you actually look at coding, you might know that Gemini is great for front-end and Codex is great for back-end. If you're vibe-coding your project, you probably want to use both, and you don't want to switch between 2 CLIs all the time. It's just a pain. So being able to use Cursor as a single way to orchestrate all the models is valuable.

Similarly, for creative tools, we're seeing this specialization and fragmentation. Midjourney and Krea, with their Krea 1 model, are the most aesthetically opinionated models. They create this incredible, beautiful imagery.

Conversely, if you look at Ideogram, Ideogram is often used by graphic designers. It is intentionally not opinionated from an aesthetic perspective. If you're somebody who's working as a creative at a big company, sometimes you're doing graphic design and sometimes you're just doing beautiful photography for print ads. You want to actually have access to both, and to do that, you use an apps company.

I think we massively overestimate the durability of revenue of AI companies more broadly as well. I think there's a chance that Cursor loses half of its revenue this year through cannibalization by Claude Code. I don't know anyone who's not moved to Claude Code. When I hear that someone's still on Cursor, I'm like, wow.

Yeah. I think the thing that we are underappreciating is that we assume efficiency is increasing, but ambition and the number of customers are staying fixed. I think this is one of the incorrect assumptions that keeps getting made around AI. It's like, well, what are all the people going to do? Where will all the jobs be?

Our ability to be ambitious and want more things always grows so much faster than our means. In the same way, if you look at software, the desire and demand for software, both to make it and to consume it, is dramatically more than the supply that we have today.

I think there is a developer and developer-adjacent archetype for whom Cursor is going to be perfect. Codex as an app, Codex as a CLI, Claude Code—all of these products are going to find market fit and all grow. If you look at any of the other markets, like creative tools, they're going to specialize and fragment in their own directions.

So when you think about market composition for that market, in the developer tooling space, does that look more like cloud, or does that look more like Uber and Lyft?

I don't think it looks like Uber and Lyft. I think Uber and Lyft are, to my mind, the most extreme examples of pure substitutes, and a lot of the price has been competed away.

You look at cloud, and you sort of have this oligopoly where they all actually have pretty reasonable margins. You can squint and say, of course, they have their specializations, but they're roughly substitutes, and yet they've all done well.

I think the foundation model companies look a little bit like that. In the apps layer, you're just going to have people who want to consume the code they generate through a rich IDE and those who want to be closer to the metal. That's probably closer to AWS and Google Cloud than it is to Uber and Lyft.

So when we think about that, how do you think about competitive investing? It seems to me like it doesn't matter anymore. When I started, it was a big problem: you didn't invest in competitors. Now everyone is investing in competitors.

Yeah.

Are we in a world where that no longer matters?

I mean, when you think about a firm that's organized the way we are, which is that we actually do stuff for our companies, it becomes very difficult to invest in directly competing companies because then you've got the same resources, the same Fortune 500 buyer, and the same engineer that both companies want to hire.

I just don't think we can run our business by investing in directly competing companies. Now, with that said, I think we're in a part of the market where companies are diverging very rapidly. So even companies that appear to be directly competing today tend not to be competing in 12 or 18 months.

Going back to what you said just before about the opportunity in the apps layer, one threat that's often posed to the apps layer is the models themselves providing products. Whether it's OpenAI focusing on health now, or whether it's Claude Code, or actually, I saw Anthropic do some Claude adaptation for legal yesterday.

Mhm.

To what extent is the invasion of the apps layer by models a credible threat to the verticalization of apps?

Yeah, so this is such an interesting topic. Granola, which we're not investors in but I admire a great deal, is a great company. They've built a really interesting thing, and they were first, of course, to do live meeting recording and transcription, which is awesome. They have been copied to the moon. Right now, everybody has a meeting transcription feature. OpenAI released one within ChatGPT. Very cool.

The thing about Granola, and I assume this is true, is that their vision is not to be a meeting transcription product. I assume it's to be a productivity suite. They're going to build Word and Docs and spreadsheets and all of these other products around that core primitive.

Does OpenAI have the prioritization, the resources, and the ambition in that direction to build all the feature surface around the primitive? I think the models will often actually recreate the primitive and even do product marketing, which I think the Claude legal stuff was.

But if you have a market that demands a lot of feature surface, I just think the model companies are less set up to prioritize it.

Do you not think that if it's bundled into an existing solution with 80% of the features, the majority of people just go, “Ah, fuck it”?

Perhaps. I just think that the model companies have ambitions in so many directions that it's hard for them to prioritize building opinionated UIs for the legal community. I also think in many of these categories, again, being multimodel is important, and OpenAI is only ever going to give you OpenAI models. Anthropic is only going to give you their models. Same with Google.

So if you are multimodel with a rich feature surface, I think being an apps company is better.

“Boring wins” is a statement that you said to me before when we were talking about the apps layer and where value will accrue. Huh. What do you mean by “boring wins,” and how does that translate to the next generation of iconic companies?

Oh, did I say “boring wins”? Yeah. Well, let me make the exact opposite case: I think weird wins.

Huh.

Yeah. Here is something that's actually very interesting: the nature of these models is very different from the nature of any technology we've had before.

I'd say a lot of the technology we've had before is quantitative and sort of clinical. It can do incredible things, but it's bounded in the range of feelings that it can capture. Now, we have this wild, non-predictable, emotional, very human technology, and sometimes it gets pointed in directions that are very human but perhaps uncomfortable to a big corporation.

The human experience often involves disagreement, persuasion, sexuality, and we see that mirrored in some of these AI products. Yet if you're Google or Apple, you have 1,000 committees that are explicitly designed to ensure there's never any persuasion, disagreement, or sexuality expressed in your products.

So I think there is a pocket where startups can really thrive, which is building these weird products that touch on many core aspects of humanity that the models can reflect but the big corporations are uncomfortable with.

What's an example?

Everything in companionship, right? Every product in companionship has been both well received by customers and a little uncomfortable for the labs to build. Perhaps even Grok.

I'm sorry, when you say companionship, you're saying, like—

likely Character.AI, but also likely Janitor AI, right? There's a ton of products that are there, like Replika, which is probably one of the most healthy and nourishing forms of companionship. All of these products are there to facilitate friendships between people and technology, and a lot of that stuff is just uncomfortable for big tech to do.

Would you encourage your children to use them?

Absolutely. In fact, one of the products that I would love to exist—my request for a startup—is what I call a contextual companion for my son, who plays Minecraft.

My son plays Minecraft. He plays it online. He absolutely loves it. The other kids playing Minecraft may or may not be the best influence—often, they're not the best influence. I'd love to actually have an AI companion play Minecraft with him, so there's a context in which they interact, and it models pro-social behaviors while still being cool and chill.

I think there's a lot of room for teaching through these types of relationships, and technology can help provide that.

Do you not think that it engenders or removes the ability to interact with other humans and makes people even more withdrawn, or used to building a relationship with technology, than we already have?

Yeah. I think it does the exact opposite. I think people are able to be more self-reflective and explore aspects of themselves and human relationships that they often just don't have another person to explore these things with.

If you're wealthy and perhaps educated, maybe you're interested in therapy, and that's an outlet for it. Or if you're like you and I and you've got this embarrassment of social riches, you have all these people who want to hang out with you, and you go to these dinners where you stay up late having all these philosophical conversations. Or perhaps if you're one of the relative minority today that is spiritual or religious in some way, and it's very emotionally nourishing to you, there are directions in which we can explore these things.

But I think for the majority of our society today, they just don't have an outlet, and I do think technology can be that outlet.

I like the idea. I also like it especially when you think about the amount of old people who are alone and you think about the companionship there.

Yes. By the way, I think the whole thing is that there's got to be a level of indirection. This is why I think contextual companions are very powerful, because I think for a senior citizen, it's important that they have a big sense of self-respect.

So, if an AI calls them every night to check in on them, they're going to be like, "Well, hold on. I don't need that. I don't need to be babysat." But if instead the AI called to check to see if they'd taken their medicine, asked them how their day was, maybe lightly flirted with them, or talked about World War II, suddenly they've got a context in which they're interacting.

There's a level of indirection, but the thing that's actually delivering is spiritual nourishment.

5. The death of the Chatbox? Why browse-based interfaces are still preferable

How does the UI paradigm change in the world of AI? This is the shittest question ever, but after 10 years, I'm not embarrassed to ask shit questions, so go for it. Everyone's like, "Now we're just all going to be voice." Do you agree with just all voice?

I think voice is amazing for enterprise. I think that, one, dynamic UIs and, two, chat UIs are overstated in consumer. The best thinker on this is actually Eugenia, who founded Replika and now likely Wabi. She's great on this.

What she would tell you if she was here is that most people don't want to save time; they want to spend time. The products are designed by the most high-agency people in the world—Sam and Elon. For them, the optimal UI is a chat box where you say exactly what you want and, voilà, there it is.

But for many people, they're looking to waste time, spend time. They want a browse-based interface. They're not quite sure what they want, and they can't always articulate it.

So, I think that in a world where we have intent-based and browse-based, browse-based largely stays the same. Perhaps the future of intent-based is chat, but I'm still a little skeptical.

People are consistently concerned about—we mentioned it earlier—defensibility, switching costs, and durability. When we think about moats and Alex's statement of "hostages, not customers" in a new world of AI, do we just accept that there's no defensibility, or are new moats created?

I think defensibility still exists and still matters. Networks are the gold standard, and they still are. A network-effect product is incredibly powerful.

Now, look, you might argue that something like Moltbook is a new type of synthetic network that perhaps means there are certain types of networks that are less defensible than they once were. But something like Airbnb—you can have all the vibe coding in the world, and its network effect is incredibly powerful.

So, one, I think defensibility matters. Traditional moats still do matter. I do think that within moats, like systems of record, there will be some that are more or less prone to disruption.

If you're an on-prem database, and there's no engagement layer and not a lot of human workflows built around the so-called system of record, I think that you actually are at some risk. If you're the core system for a bank, you've got thousands of transactions per second, hundreds of humans that interact with you, and this incredible demand for accuracy, I still think that you're as good as gold in terms of defensibility.

Are there any forms of defensibility that were very prominent in the prior 10 years which are no longer as prominent?

Yeah. I'll give you the opposite. I was always skeptical of the sort of data network effect. That was the thing that got thrown out a lot when you couldn't think of what moat to say.

But today, if you look at companies that have proprietary data sets—not just proprietary; likely OpenEvidence is a good example of this—but live, proprietary data is a very, very powerful moat.

When you say live, what do you mean?

Your health data, for example, right? That's a sort of live and ever-changing source of data. Now, there's a question of how proprietary that can be.

But once you actually have data like that, or perhaps live data about a product that's running, you can put a relatively commodity model in front of it and get much better results than the most cutting-edge model that does not have access to the proprietary or live data.

Okay. So, we have relatively the same forms of defensibility that have existed before, that we'll continue to make money.

I'm always trying to understand. I feel very insecure right now as an investor because I'm trying to understand what holds true from the prior decade and what doesn't, and I need to change my mind. When the facts change, I change my mind.

When we think about a lot of the forms of defensibility remaining true, I was always taught that margins matter. I walk with my mother around London—poor woman—and I'm always like, "Mom, margins matter."

Yes. I can imagine holding your hand. Poor Mom, handheld. Yes.

Jesus. No wonder she wants to finish the walk.

My question to you is: Do margins matter as much in a world of AI, and are we entering a new way that we should be thinking about margins?

Yeah. So, here's actually where I think there's nuance in the margin conversation that's important. We should talk about the bubble that doesn't exist, or perhaps there is some sort of subsidization and distortion happening in the market. For the record, I don't believe we're in that period.

But I do think that any time you have these sorts of superheated markets, you have some distortion. If you look at the distortion from 2021, you essentially had this indirect subsidy of Google and Facebook. You would invest in a fintech company, give them $10 million, and they would go spend $8 million on Google Ads and Facebook Ads.

So, there was a subsidy happening. Those were sort of empty calories for the startup. If instead you look at the form of subsidy that happens today, what it typically means is zero-margin or negative-gross-margin credits for the user to try the product.

These things tend to be a drag, but they're actually very healthy calories for the companies, because out of that you get conversion into high-paying users, many of whom are actual power users.

So, I do think that the blended-margin story for AI-native companies tends to be worse. But if you look at the overall form of distortion that's happening, it's a much better one than we had 5 years ago.

6. Why power users are 10x more valuable in the age of AI consumption

Does that make sense?

Jason Lemkin is a very good friend of mine from SaaS, and he said a brilliant statement to me yesterday. He said, "For the best companies, influence is the new sales and marketing."

Yeah, I love that. 100% correct.

I also think that power users are so much more powerful than they ever have been. Andrew Chen used to say pre-AI—and I love this—"Power users are just users." It was true, because even if they got 100 times more value, they typically didn't pay 100 times more.

You look at Spotify, a great European company. The very best Spotify SKU, with the highest-bit-rate music, totally lossless, all the podcasts, all the videos, the family plan—everything—was $20–$25 a month. So there was a belief that the price ceiling for mass-market consumer products was $20–$25 a month.

You look at Grok Heavy, and it's $300 a month. ChatGPT is $200 a month. Gemini Ultra is $250 a month. So we're seeing 10× higher prices being paid, and you have consumption revenue on top of it. For power users, they're paying incredibly high subscription rates plus consumption revenue, so the S&M costs of acquiring those users are very wisely invested.

7. Do margins matter in a world of AI?

But you're telling me, then, for my team, when I'm looking at margins with the investing team, that we should have the same high bar that we carried, or that we should have greater elasticity to lower margins?

So, first of all, it's typically a lot of organic traffic. One, I would look at your M1—your sort of month one—as traffic, not truly acquired users, because it's organic and it's free to acquire. Second, I would take a look at the margin cost of those users' free trials and just say, “Hey, that's CAC, and that's okay.”

Then look at the margin profile of people who convert and say that's the durable margin profile of the product and the business. Does that make sense? You're unbundling the CAC-oriented margin spend versus the durable margin, which is what's associated with your power and paying users.

It totally does. The challenge becomes, if you're trying to work out a CAC-to-LTV metric that you can oscillate around—

Yeah.

—it's very difficult to get an accurate sense of LTV in such a changing landscape, where you're not sure of the durability. Is the LTV 12 months, or is it 46–48 months?

I think that retention really matters. If you take a look at the best AI products, even if you look at M2 as the new M1, because, again, you're getting a lot of tourists who come in at M1 and you're not paying anything for them—

And so M1 means month one.

Month one, that's correct. If you look at M2 as your first month for some of these products that are acquiring a ton of top-of-funnel traffic, then you apply the same high-retention bar you ever applied to them. What's an M12 that would make you very excited?

I mean, the bigger the better, but certainly 50% is solid, right? If you're at 60–70%, we're very, very happy.

8. Why we are definitively not in an AI bubble right now

Okay. So we have that in terms of margins. I do want to touch on what you said about the bubble. No, I'm not in that camp. I like to attack things while I'm there. Why are you not in that camp?

9. Lessons from Marc Andreessen: Why the "quality of being right" supersedes process

There's a little quip that I like to use, which is, “It's not a bubble, and it's good that it is.” I'll tell you why. This is not my area of focus or expertise, but, one, you look at OpenAI's recent investment announcement, which is that they're at $20 billion of topline. The way that they got there is they 3× capacity and they 3× topline. Every time they bring on capacity, all of that supply—that inference supply—is 100% spoken for.

We're seeing that story happen over and over again. Whereas in previous so-called bubble periods, you saw this incredible build-out of supply far ahead of demand. So far, we're not seeing that.

Two, if you actually look at the prices that customers are paying, they're going up. You're not seeing the price compression that you would get from a typical overbuild of supply.

Three, as I said previously, even if there is subsidization, there's always going to be some distortion and subsidization. It's a sort of intelligent subsidization that's mostly being paid for by Big Tech and the labs, and it benefits consumers and startups. God bless. I'm all for that.

I do a show with Jason Lemkin and Rory O'Driscoll, and Rory said something brilliant, I think, which is, “This will all work out if we see the transition of spend from the 12% SaaS budgets that we operate in today to the human labor budget.”

Mm-hmm.

I mean, we're already seeing it. I think D.G. was on the show talking about C.H. Robinson, right? We're seeing a lot of companies start to see the productivity improvement from this new technology. How can they not?

It's not just coding agents in which this is showing up. You talked about voice. Voice is the wedge into the enterprise. Voice agents are so powerful. By the way, I think the near-term story of a lot of voice—we talked about support, and you talked about customer support—that's interesting. But the more interesting thing is, why is support an isolated function?

Let's go through it. Typically, you've had sales, support, operations, and collections. Who is the person who's really good at customer support? They're empathetic, they're a listener, and they really understand the product well. Who is really good at sales? They're more of a yapper. They're a talker, they're high-energy, they're very charismatic, and they're good at the upsell. They're always in a good mood.

You've got these 2 different human archetypes for these 2 different roles. We've typically organized the enterprise around these 2 archetypes, right? But now the models can be either of those people at any time. So the most sophisticated companies are starting to take support, sales, collections, and operations and bundle them all together with 1 broad goal, like CAC improvement.

10. Why the Legal & Customer Support industries will have dozens of winners

I think that is going to be the 10× on productivity, more than saying, “Hey, we're just going to take cost out of customer support.”

How do you think about competition within markets? I'm jumping around so much, but I'm just fascinated. You brought up customer support. I tweeted about it the other day, and I've tweeted about it before. I just can't get my head around this market.

There are about 50 providers with over $50 million in funding, and 10 with over $100 million. I was very much of the Peter Thiel school of thought that competition's for losers, and we want to have monopoly markets, with your Decagons and your Sierras and your Intercoms and your Palonas. I can go on and on.

I don't know. Well, the question is, how do you define a market? This is an important point. I would argue that, in many cases, what you're calling a market is actually an industry.

Let's look at legal. Many great companies have been funded, and there's still room for another dozen. I think the reason for that is legal is a $500 million. It's sort of infrastructure for capitalism broadly. Is there going to be 1 company that wins the entire market of infrastructure for capitalism? Of course not. That is an industry, not a market.

11. Why the developer tool market looks more like Cloud than Uber and Lyft

You're going to have dozens of winners that all specialize, just as in legal today, you've got dozens and dozens of specializations. I think in many of these markets, we're talking about them as if they are 1 market, when they are much, much bigger, and all the companies will specialize in their own directions.

Well, it's a $500 billion market if you assume that we eat their market, not that we're an attachment to it. Correct?

Yeah.

And we aren't attached to it.

I mean, that's an open question. I don't think that we're in the 8–12% anymore, right? $50 billion in legal software traditionally. I think we're going to be somewhere between the $50 billion and the $500 billion, and I think closer to the $500 billion than the $50 billion.

What does that look like? That means AI-native law firms?

Possibly. I think it means dramatic productivity increases for lawyers and dramatic productivity increases for programmers and engineers.

I think the difficulty of doing 100% of a job is really, really high. It's pretty easy to get to 60–70–80%. So I do think that's why a 20% productivity increase, so far, we're seeing it show up more as a 4-day workweek than 20% fewer jobs, because jobs as bundles of tasks don't set themselves up to be 100% automated so far.

You can do all the customer support you want, but sometimes you've got to take the customer out for a steak dinner. So far, the models are not doing that.

They're not. We mentioned the $500 billion TAM. Do you do TAM analysis work when investing?

Here's what I think. We tend to consistently underestimate how big the markets are and consistently overestimate how easy it is to go from 0 to 1.

When you squint, you can take something that's not working and say, “I can see how it will work.” That is why, in my mind, seed investing is its own sort of art. I focus very much on Series A because I believe that having shipped something and having sold something is such a dramatic signal.

To me, that is actually the optimal point in terms of information provided versus entry ownership and price, whereas at the seed, it could be anything. It's very, very difficult to get something working.

Once you do get something working, I believe these companies tend to be even greater and greater versions of themselves for a long time to come. I think the mistake that many venture capitalists have made is just not estimating the market to be as big as it is.

I'm enjoying this so much. I have really 3 things I want to dig into there. You said that markets are underestimated in size. Our dear friend Alex at Deel—I met him at the seed round, and he told me about Deel, and I was like, “Dude, you're brilliant, but payroll.”

I'm sorry, brother. Deel at $11 billion? I'm sorry. Fuck off.

I didn't tell you earlier. What's your next pass? Please give me a granular answer.

Okay, well, great.

Yeah, yeah. Chris knows this. I sent a voice note to my partner saying, “Will someone please set up a JustGiving page for Chris? No one’s going to invest.”

I will. I will send you my next pass. You know, most investors, when they send you the pass, you’re like, “Wow.” I look forward to your down.

Yeah. Mine you should do.

Yeah, 100%. So you said about market underestimation. I underestimated the payroll market specifically. I thought Alex was right. I didn’t underestimate him, but I underestimated the market. What market did you underestimate that you later realized you were wrong about, and what did you learn?

It’s such a good question: Which market did we underestimate? I’ve made this mistake a couple of times. For example, I remember when we were acquired by Google, looking at the stock price then and telling my co-founder, “Well, maybe this can go up 10, 15, 20, or 30%. How much bigger can it possibly get?”

If you look at a company like that, which was so capable but seemed dominant in its core market, it was very hard to squint and see what it would become. It’s so much more valuable than it once was, right?

I think another interesting example of this is Credit Karma: free credit scores for Americans. I know the credit score is a much bigger concept in America than it is where I grew up in Canada or even here, but you would ask yourself, if you did the back-of-the-envelope calculation, “Well, most people tend to use their credit score once or twice a year, right? And most people don’t even actually need it that often. It’s only when you’re applying for a new financial product.”

For most people, you either have exceptional credit and you don’t really need to look at it because you already know that, or you have terrible credit and you just don’t want to look at it because you already know that. So now you’ve got this cohort of people who infrequently need access to their credit score. Is that really a big company?

If you then look at Credit Karma, over 100 million Americans use it. You’ve got 50 million quarterly actives, and people log in, on average, 4 times a month. The reason that it works is that the credit score is actually this sort of mirror that people like to look in and see how they’re doing objectively as an adult—whether they’re doing great, whether they’re doing poorly, or whether they’re doing just okay.

People really find a lot of satisfaction in the feedback loop of looking at their credit score. That’s not something I ever would have predicted. As a result, Credit Karma has many opportunities to inform and sell their customers products, and it really, really works.

So how do you reflect on missing that?

I think when you have a formidable founder and they’re showing a lot of early momentum in a market, inertia is the best mental model. In my mind, inertia is the most powerful force in the universe. Everything that is happening today is going to, by default, happen forever.

When you have a formidable founder making tremendous nonlinear progress, you have to tie-break in the direction of them doing it forever. That has to be your underwrite.

It’s so funny. Roy just mentioned earlier, he says, “When a founder continuously hits target, you should bet on them continuing to continuously hit target. Don’t overthink this. It’s hard to hit target.”

Well, this is—I’ll tell you a funny thing. When I first started, I spent a bunch of time with Marc, Chris Dixon, and everyone. I remember sitting down with Marc and saying, “All right, Marc, what’s the process? Tell me exactly what the process is.”

Marc said this maddening thing, which is, “Just be right a lot.” I was like, “Of course, be right a lot, but what else?” There were a bunch of things that we talked about, but ultimately, having reflected on that, I think his view is that your process doesn’t matter as long as you’re consistently winning.

When I started my career at Amazon as an engineer in 2003, they had a very similar thing—I think it’s still a part of their leadership principles—which is that you’re consistently right. I remember being 23 or 24 years old and finding it maddening because, well, why are you right? How are you right?

But this quality of being right sort of supersedes the why or the how, or all of our very intellectual mental models of how long it can sustain.

You said, “Just win,” and the importance of winning. We said downstairs, I lost to a wonderful colleague of yours, Seema, in a company, AskLio, in Germany, at the Series A, and I reflect on this a lot. A lot.

I can tell it’s on my mind.

When you reflect, what was your most painful loss, and how do you reflect on that?

I haven’t lost a deal.

You’ve never lost a deal?

I’ve never lost a deal.

How long have you been in Andreessen?

6.5 years.

Huh?

Yeah, yeah.

Do you worry about that? I mean this in the nicest way. I asked Ravi Gupta about this because he lost the A of Rillet and then did the B, which is great and fantastic. Well done to him.

Yeah, perhaps the risk aperture is not high enough if you’re never losing deals. Maybe. I don’t know. I think there is a process by which you can be a part of most important companies that you want to be a part of.

I think there are some very difficult pre-existing conditions to overcome, like somebody having a very healthy, successful relationship with a past investor.

You’re just never going to overcome that, right? And, by the way, having been that healthy and supportive past investor for many people, I would never expect those founders to go work with someone else. What do you do in those situations when they’re like, “Listen, I love you, but I’ve known these guys for 10 years. They backed me before”? Do you say, “Hey, we’re going to be the collaborative partner and try to nestle in now,” or do you just peace out and not take part?

I think that there are no games to be played. I think this is the magic of being in this business and being at Andreessen Horowitz.

When I started, I had this nervousness around, “Maybe it’s a sales job,” but I’ve realized that if you just show up with the right intentions, you have to assume that they know everything. Of course they do. We live in an era of very, very sophisticated individuals and founders, and you respect that and say, “Look, I want to respect the relationship that you have.”

With that said, our mission is to be a part of every important technology story that happens. If there’s a way to be a part of it now, great. If not, let us get to know each other and earn the right to be your lead investor at the next round. Sometimes that’s the right thing.

How elastic will you be on ownership in order to win deals?

Not very elastic. I try to explain what our model is, but I’m very elastic on price. I should probably be careful about saying that.

I want to learn from you.

Yeah, yeah. Well, I mean, this is my mental model.

No, no, it works. I’ve learned in venture that simply copying often works.

Very elastic on price. Below a certain price, it doesn’t really matter. What is that price? I mean, look, I think price starts to really matter once you’re into the hundreds of millions, and certainly at the growth stage, price really does matter.

But I think at the early stage—let’s say sub-$100 million, like $50 million, $70 million, $100 million, even $120 million—the main way that price shows up is that it may impair your ability to raise the next round because you priced something so high.

We’re very transparent about that. I’ll tell a founder, “Look, you’ve got great metrics. We can do this Series A as 12 on 60, 15 on 75. Twelve to 15 is a little bit of a wash for us in terms of the check size that we’re writing, and it’s more about what expectations you want to sign up for at the next round.”

The one thing that we typically don’t flex a lot on is ownership, because that is our whole model. The model of, “Hey, we’re going to put all the chips in behind you,” doesn’t work if we’re not real partners.

200 versus 300?

Yeah, I mean, it’s in the margins. Again, I think a lot more at that sort of price threshold. I start to think a lot more about the next round than the absolute dollars in, right?

The absolute dollars, again, for a sufficiently large fund probably aren’t going to make or break the fund, but your ability to raise the next round—especially once you’re in that growth territory—is important. The $500 million round is a hard round. The difference between having to raise at $300 million, $500 million, and $700 million is pretty significant.

Do you think we’re skipping that round? If you think about the companies that are raising at $100 million to $200 million with $1 million to $3 million in revenue—say, early signs of product-market fit—and then they’re growing so fast that they’re hitting $30 million to $50 million within, I don’t know, 12 to 24 months, then they raise $1 billion and that $500 million in-between round is now gone.

12. Is "Triple, Triple, Double, Double" dead?

Yeah, that’s right. Look, I think that happens, and in a case where it’s because core metrics are super healthy, good for them. God bless. I have a lot of enterprise SaaS companies that do double-double or triple-triple-double-double.

Is the world of triple, triple, double, double dead, and do we all have to be Lovable, Replit, and ElevenLabs to get funded?

I don't think so. A lot of it depends. It's calibrated to your part of the market: product velocity plus business velocity. I do think that you have to be top-quartile compared to your peer set.

I think there are some markets that are consumer-led or bottom-up, where you can just see this explosive growth, and that is awesome. It's extraordinary to see. But look, if you're selling an ERP, you've got a much more cautious customer. It's a much more high-stakes sale. If you're selling payroll—now, granted, in the case of payroll, Alex and team have done a tremendous job of what should be a slow-boil sale and turning it into a fast-boil sale. They've got some very specific ways that they do that, but typically, that is an industry that moves on slower cycles.

So, I think there is just physics to some of these markets that means triple, triple, double, double is phenomenal, but there are other markets in which, especially with these new primitives, you can go 10 to 100 or 10 to 200.

So you bring triple, triple, double, double to partnership, and they won't shit on it?

Absolutely not. No, look, again, I think these are all heuristics that we use and throw around. First of all, I want to say that I respect the difficulty of getting to 1 million in revenue. It is so hard. Getting anyone to pay you anything that's not a family member or friend is hard. Then going from 1 to 5 or 10 is super hard, and 10 to 100 is tremendously hard.

So, one, I hate it when investors are very flip about this. And two, I think it's all about the assumptions that the founder is making, the data as a way to validate those assumptions, and the direction—the “what if it works?” What is the direction of the curve? What's the area underneath it?

Right. Totally get you. And, by the way, sorry to interrupt, there are companies that are area-under-the-curve companies, where it may be a much more complex, slower-growth story, but the area under the curve is much more significant than companies that have a very high slope but potentially have challenges with defensibility.

Absolutely. Yeah, area under the curve. What do you have today with Figma? You have an N-of-1, network-effects product, right? That, by the way, is ahead of where I think the market is going in terms of moving from products focused on execution—which today are being subsumed by coding agents—to markets focused on thinking, right?

I think a lot of the thinking work is going to be done in products like Figma. I'm not sure that Dylan and team saw that 10 years ago, but I think they're well positioned today.

Area-under-the-curve companies: How does the world change for them today? Do they still hold inherent value for fundraising early? How does that change because it's difficult to sell?

I think the challenge for area-under-the-curve companies is that you've got to have enough momentum that you can continue to fundraise. You've got to have enough substantiality that your customer loves you. They're willing to pay you upfront, and they're willing to expand with you.

So, it has its own idiosyncrasies and difficulties. But I think often, some of the most significant companies are these area-under-the-curve companies. There are these 20-year overnight success stories, and I think those are underestimated in venture lore.

The 1-to-100 companies are extraordinary, and we all love to be a part of them, but we may talk about those and lionize those perhaps sometimes at the expense of the area-under-the-curve companies.

You said you very much focus on the Series A. We mentioned some of the pricing differences there. I get in trouble with my team constantly for tweeting things, and they're like, “Oh, Harry, you make my life so much harder.” And I'm like, “It's too easy otherwise.”

I say that Series A is the hardest place to be investing right now because, essentially, you have 1 million in revenue and very little sign of product-market fit, honestly, at 1 million in revenue.

I disagree.

You're paying 100 to 200x, and it's incredibly competitive. The price-to-progress ratio is incredibly mismatched between a $25 million seed. Why am I wrong?

Well, okay. First of all, I think as an investor, you have to decide what kind of risk you want to take. That's what we're paid to do. So, let's talk through the different types of risk.

The first risk is one that you mentioned, which is competitive risk: Can I win this process or not? The second risk—and this is maybe a slightly less good risk to take, but I think still a fine one—is pricing risk: Did I overpay? Again, I think the way that shows up is the difficulty of the next round based on your entry price, right?

Maybe the third risk is team risk: Can this team actually go the distance? Can they be big enough to fulfill the company's ambitions? Because I think the founder themselves can attenuate or amplify a company's destiny, right? The company can't be bigger than their ability for it to be big.

The fourth one, I think, is a little bit of geographic risk. Maybe this is less true today, but is a Silicon Valley team going to do this? And the fifth is fundraising risk. This is a non-consensus deal that actually has no other investor interest around it.

That is not to say that you need investor interest, but if the team has a difficult time fundraising, no matter how good their product and technology are, they're not going to get the opportunity to see their vision through. So, I think taking competitive risk—can I win?—and pricing risk—

When you say, “Can I win?” you're saying, “Can I win as an investor winning the deal—”

Against the other VCs. That is what we should do. That is the number-one most important thing that we do: win the deals by building trust with the founders, being smart on the markets, being first to conviction—all of these things. That's why I think the Series A being hard—it's supposed to be hard, and you should be winning anyway.

A couple of questions on the back of that. They're leaving their startups faster than ever, having just raised big rounds to do new things. Are we seeing this increased promiscuity from founders, do you think?

I think we've always had to assess how authentic their connection is to the problem at hand, right? Because doing these startups is a little bit irrational. Alex said this on the pod, and I think he's exactly right: You have to be a little bit irrationally optimistic to do it.

I think you also have to be irrationally interested in the domain in which you're working, because these things get hot and cold all the time. I think that authenticity—which is not a comment on intent. Sometimes really well-intentioned people—I've been this person—have a reason that they're building their company other than an authentic connection to the problem.

I just don't think that's a good setup. That's a setup for promiscuity.

Do you mind when someone comes in and says, “Listen, I don't have any particular interest in sales for car dealerships, but I saw it as a ripe area for innovation and where models can be transformative”? Do you mind that?

I think there has to be some sort of irrational direction in which they're pointed. Maybe it's pure capitalism, and they're like, “Look, I've studied the living shit out of this market, and it is a means to an end for me, but I'm going to get there or die trying.”

You need to see a little bit of that outlier emotion and commitment. I think it's best expressed when it's in the direction of the problem, but it doesn't have to be. I think if somebody comes in and they're like, “I did a case study on it, and it looks great,” that's not a great setup.

Do you find with the founders that you work with that the best founders are the best fundraisers, or actually can they be a bit quirky? Are the best founders generally great fundraisers?

I don't think they all have to show up the same stylistically—the polished, go-to-market-oriented, whatever, the type of founder we saw more often 5 years ago.

The Krea guys, to me, are a great example. They come into our first pitch, our first meeting, with everyone in the room over Thanksgiving holiday, and I think they're both wearing matching kimonos. They're both drinking Celsiuses. Victor's got his long skater hair. He's just this total badass who looks exactly the opposite of every MBA founder that we had been meeting 5 years ago.

He's got a quiet presence, and a lot of it comes from his command of the technology and the domain. It's a totally different style. It works really well for fundraising. So, I think you do need to be able to fundraise, and you can be very authentic in the way that you do it.

In terms of having a command of technology, and you said earlier about the challenge of shipping product and getting from 0 to 1, showing that you've had success building in the past is a great way to prove that you can do it moving forwards.

Do you have an unreasonable or unwavering leaning toward serial founders who've proven that they can do it because of their track record?

Yeah, I'll give you a nuanced take on this. I think that repeat founders working in their domain of expertise are formidable. The Clutch guys sold a company to Carvana. They weren't super happy with the way that the whole thing ended up, in terms of their startup achieving their ambitions. They then went and started another company out of that, also in the auto space, called Clutch. It's going extraordinarily well, and they know they're taking all the shortcuts because they know the market.

So I do think, particularly in enterprise, working in the same domain and being a repeat entrepreneur is a huge source of alpha. I actually think, conversely, in consumer, having a beginner's mind and a high willingness to be embarrassed is a competitive advantage, because so many consumer products feel embarrassing and are immediately dismissed as embarrassing or impossible or a silly, non-serious thing to be working on.

When you're 25 and the stakes are low, you just want to make something happen in the world. That is a perfect setup. Once you've sold a company, all of a sudden your venture friends are like, “What are you working on?” Your girlfriend or your boyfriend is like, “What are you working on?” You want to sound cool at dinner parties or at the bar. That slight hesitation to be embarrassed can sometimes hold you back from the most ambitious, interesting consumer ideas.

As I said earlier, when the facts change, I change my mind. What about the way that you used to invest has changed most significantly?

Well, I think the number one thing—and here's my free advice to other investors, but also founders—is that you have to use the products today more than ever.

I think the investing landscape of 5 or 7 years ago, when there was a ton of fintech—and I'm a fintech guy, I love fintech—made it harder to build intuition for a small-business factoring solution. Maybe I should start a small business just to factor, but there are too many steps to actually try the product. Today, being native in this product cycle just means waking up every day and being like, “If there's 3 new models today, I'm going to try 3 new models. I'm actually going to make something.”

Holding yourself to an incredibly high standard of trying everything just gives you so much information and intuition. I think it's nonnegotiable for founders, and I think it's incredibly important for investors as well, yet most don't do it.

When you talk about trying products, 90% of the companies that I get pitched, especially on the application side, say they're agent-led or agent-first. You said before there might be agent overhype. Can you talk to me about this? Why do you feel there's agent overhype today, and what does that mean?

Here's what I think. I think that the extremist view that we're going to have autonomous agents that simply do everything over incredibly long time horizons—maybe we'll get there someday. But I do think that, at a minimum, you need humans in the loop for exception handling.

These models are only as good as the instructions that we give them. Our instructions—think about the way you manage your team—are often frustratingly vague. So I do think that we need people in a tight loop with the models to actually achieve our objectives. I think that the sort of agent-maximalist view, which is, you just chill out for the day and your AI does everything you need it to do, is probably a little bit ahead of where we actually are.

Do you take the view that agents won't remove tasks from what you do, but they'll enable you to do tasks that you didn't have time for?

I think it's both. I think that they will do tasks. They'll do the low-NPS work that you don't want to do, right? Do the work that you want to do, not the work that you have to do.

I think the second thing is, yes, the surface area, the sort of circumference of ambition, is going to go dramatically up for us as individuals, but also for us as a species. Harry, how can this be the peak expression of our ambition as a species? You've read enough sci-fi books to know, or even have a glimmer of what that looks like.

I think that's a world that we're going to actually live in, which is, if you are ambitious in a direction, you should be able to fully chase it down and express and fulfill it. The only question is, who are the ones that are ambitious to go do things? I don't think execution or expertise is any longer a constraint.

When we think about the best places to win an agent-first game, I had Eric from Podium on the show, which is a fascinating story of a traditional SaaS provider that's now got a $100 million-plus agent-led business. He said that, fundamentally, if you want to win in an agentic world, you have to own the tools, the workflow, and the data.

You have to own the full stack, à la, of course, him and Salesforce. Do you agree that you have to own the stack to really be a big player, or can you be a meta-layer on top of a core provider?

Yeah, I mean, I think that you can use a core provider via tool use. To me, ambiguity is one of the big questions around how much leverage you get out of agents.

If you look at BPOs—business-process outsourcing—they are the areas in which there's the least ambiguity, where the job is literally a series of tasks, where people in offshore call centers take a task off the queue. Those things are very well set up for automation and agent replacement because you've got incredibly well-defined tasks, and you've got jobs that are bundles of these well-defined tasks.

I think there are many jobs in which there's just such a high degree of ambiguity. Even in software development, arguably the coding is the easy part. The tough thing is, what are we coding today? How do we adjust and how do we adapt to what the customers are saying, what the market is saying, and what our own epiphanies were overnight?

The models are incredible at getting us into these local maxima, and sometimes the local maxima is the global maxima. But I think often it takes human intuition to break from the local to the global, and that is something I think the agents just aren't going to do.

You mentioned BPOs there.

Yeah.

How do you think about the future of UiPath? They've had a tumultuous journey in the public markets. Is that one that's sadly suffering from this, or are they actually well positioned to take advantage of the distribution that they have?

13. Open vs Closed Source

I wish I knew more about UiPath. I just don't know enough about the company to comment. I think RPA is super interesting, but vision models haven't nearly kept up with the way that we've talked about them.

Can I ask you one thing that we haven't discussed, which I want to? It's kind of open versus closed. It's a big question. I find a lot more people use open than they actually say publicly, and there's a lot more willingness than ever to use open source.

Open source.

Yeah. How do you think about the distribution in terms of open versus closed, and how does that look in the next 24 months?

That's a good question. I don't think we're at a point in the cycle where companies are focused primarily on cost optimization. I think that is one of the reasons to choose open, which is, get an open-source model, host it, and then have a cost benefit as a result.

I do think there have been some interesting properties of open models, like Kimi K2. I believe that they didn't post-train it to constrain what it could say. As a result, it was just a lot more interesting in terms of text generation in many directions. So it had this sort of interesting product characteristic that a bunch of companies built around—a bunch of companion companies, in particular.

I think there are these idiosyncratic reasons we choose the open models for product quality, but in most cases I think companies are thinking about maximizing the direction of ambition and their ability to fulfill it versus taking cost out, and closed is still a bit advantaged there.

The nice thing about closed is they, too, have been cutting their costs, right? Granted, closed is more expensive than open in many cases, but the cost of a token on GPT-4o has gone down 100x since the model was released.

You said that we're not in the period of cost optimization, which I agree with and think is a very interesting point. Jason Lemkin said to me—he's a real builder with one of your tools, actually, with Replit. I think he's literally one of their top users. It's insane what an incredible power user he is.

But he uses ElevenLabs as part of the voice for one of his games. He said, “This will be the year where we see the true substitution of AI products based on price.” He's like, “ElevenLabs? I love it. It's amazing. It's too expensive. It's too expensive. This is the year where we've moved from trying things to [__] it works, but it's too expensive.”

No comment on ElevenLabs. But do you agree that we're going to see this transition in mindset from “[__] it works” to “[__] it's expensive”?

I don't think so, because what we keep seeing is that, as the models get better, downstream players' ability to take those capabilities, productize them, and raise prices has outstripped the rise in costs, right?

The incremental cost increase—potentially, and in many cases not a cost increase, but not a cost decrease—is so far outweighed by what the new capability unlocks. Look at coding agents. What you can do with coding agents—Claude Code came out last February—is dramatically better. Is anybody here saying, “Well, I should go back and use Claude 3.7 Sonnet because it's cheaper,” you know, or, “I should use something other than Opus 4.5 or Codex 5.2”?

Nobody is saying that because the capability is so much more powerful. It really sparks your imagination in the other direction: What more can we do, rather than how do we make the existing thing cheaper?

It is interesting. I remember when he said he's got a startup game, and he's like, “Dude, I'm terrified that people are going to use it because I'm going to go bankrupt.” I was like, “You okay?” But this is actually a fun topic, which is that the fact that these products have costs is a very good thing.

What it means, then, is Jason's going to have to figure out his business model early. This field-of-dreams investing, where a company builds a free product and they're like, “Someday we'll figure out a business model,” is not viable. It's like, no, you have costs today, the way that every small business in the history of small businesses has had pre-software. So the fact that these companies have costs actually forces a business-model hygiene that I don't think existed across the board 10 years ago. And that's a good thing.

If you were ElevenLabs, would you not just say, “Fuck it, subsidize, cut prices, own the market. This is a land grab. Don't risk churn by raising prices for the next year or 2”? They just raised $500 million. They could have raised $5 billion more.

I think, in general, there's so much more to be done at the frontier, and there are so many more categories and capabilities that those things unlock. It's just a better use of time. Today, again, we talked about 8% to 12% of enterprise spend being on SaaS, right? How much of consumer disposable income is spent on software today? A few hundred a month, maybe.

We're going to asymptote to 80% to 90%, I believe, for consumer spend and enterprise spend. The way we do that is by pushing the frontier, not by—

80% to 90% of consumer spend? The sort of discretionary spend—of course, there's going to be, you know, software—

Rent and food. But yes, dude, I think that software is going to eclipse many parts of our discretionary spend. We just talked about companionship and friendship. We talked about entertainment, potentially therapy, potentially health care, potentially professional, right? A lot of the spend that I do on things that help me be better at my profession—education.

So there's a tremendous area for software to expand into. Let's forget about taking cost out of things for now.

I don't know if you're including food in that. If you're including DoorDash, they can get on board.

Discretionary, nondiscretionary spend is fixed. And you're clearly not European because you missed 1 crucial one, which is fashion—

Which would not be that—

You're going to say wine, and dude, no one buys wine. No one drinks anymore.

Okay.

Tough to be in the wine.

14. Is Kingmaking Real?

Not even the Europeans. No, no, no, not at all. It's super interesting. Do you agree with kingmaking today, in terms of the belief that there's an anointed winner? Do you think kingmaking is real?

I think 1 example of where there is a very positive catalyst in the investor base for enterprise companies is YC. YC is an awesome place to start an enterprise startup that sells to other enterprise startups, and they've got these good vibes within the community that make it easier to sell into even much bigger, more established YC companies.

So I think that's a good example in which picking the right investor is actually a big benefit. A lot of what we do is connect companies that are small but have really important product and technology to the Fortune 500 and 2,000, but we can't force them to buy that technology, right? And again, you have to assume that the buyers, especially these days, have perfect information.

I think that the right investor can be a catalyst, but I don't think that you can take a product that would not otherwise be the winner and anoint them the winner.

Do you agree that the best founders you work with don't need their VCs?

I think the best founders that I work with know how to maximally leverage their VCs. And look, I think there is a set of founders who perhaps would never need their investors, but I do think that the best founders know how to extend their success and increase their momentum by leveraging the right investors, like Alex does, right, dude?

I mean, I basically have a sales quota with Alex, and D.G. would say the same thing. And Ben—he's even calling Ben, saying, “Hey, Ben, can you help make this introduction to XYZ?” He knows how to get the best out of Andreessen Horowitz.

It's not just the investors; the entire team shows up for him that way. Could he do it without us? Of course he could.

What advice would you give, then? We have so many founders that listen. What advice would you give to founders on how to have maximum value extraction from an investor base?

Well, first, pick an investor that does stuff. I think, number 1. I think number 2 is—

How do you know? Everyone says they do.

The best way is to talk to other founders, right? You should talk to other founders. I think the second thing is, again, the VC can't distort the market, right? All they can do is make all the introductions.

And I think the best thing—Marc talked about this—is when you're small, the VC sort of gives you power, right? That's what you want. The VC basically takes your brand, which is not big, and they lend you their brand. So you're not XYZ company; you're an Andreessen Horowitz company.

Now, over time, your brand becomes much bigger than Andreessen Horowitz, and that is great, but they can help bootstrap you and create credibility in conversations. You still have to have the best product, technology, and go-to-market to go win the customer.

I totally agree with that. I think the lending of brands, I think, is how you've described it before, is phenomenally valuable. Can I ask you, when we think about the lending of brands, who's the single best founder you work with?

Alex is just such a beast in terms of his go-to-market instincts, his product creativity, and just his responsiveness. The guy's nuts. Alex is 100% working all the time. It's just incredible.

I've got a fun story for you. We have Project Europe, which is like the Thiel Fellowship for Europe. It basically backs 18-year-olds with a big dream and technical capability.

Oh yeah, he was telling me about this.

It's amazing. Anyway, I pinged Alex on a Sunday morning at 7 a.m., saying, “Hey, they want an intro to a sales rep on your team. Who's the best person?” He's like, “Intro to me, please.” I'm like, “Dude, it's like a $1,000 deal.” He's not worth your time. He's like, “No, no. To Alex at Deel.com.”

That's what I mean. Yeah, he's so impressive.

But look, there are other founders who have specialists in their domain, like the Clutch guys I mentioned, who are deep technologists and know how to apply it to product, like Krea or the HappyRobot team. So there's just so much to learn from all these individuals, and I know it sounds trite, but I'm privileged to work with them.

No, the HappyRobot guys—I wish I was invested in that.

They're amazing, man. Please don't tell me you passed on that at the seed, too.

No, no, I never met them. Thank God. Okay, good, good, good. No, no. That's one of those ones where it's like, I wish I was in it, but I never had the chance to be in it.

Incredible technologists, really earnest people, and they're seeing a ton of success.

When you reflect on companies or investments that you've made that were not good, what did you not see?

I think, again, if there's a mistake that I've made, it's been being a bit too casual about product-market fit. This was more of a 2021 mistake, which is assuming something had product-market fit when perhaps it didn't. And perhaps the founder had a super-credible theory—which, by the way, matched my theory—for why it would get to product-market fit.

But, as I said, it's easy to overestimate the path from 0 to 1. I'd say if there was a mistake I made, it was not being intellectually honest about whether this was actually working or whether I thought it would work in the near future.

Now, look, I've done a bunch of seed investing and I've made the bet, and I think if you're intellectually honest and clear-sighted about a belief that it will work, then that's a fine way to invest. But investing with this self-deception of, “Well, let's just assume it's working when it's not quite working,” is a mistake.

Do you think you know whether it's a good investment or a bad investment in the first 3 months?

Do you agree? I don't know. I think the area under the curve—companies take time to develop.

I get you.

Here's what I'll tell you. I think that there are moments when you win a deal and you're just like—the feeling is sheer relief. And I'm sure that there are moments—I haven't experienced this, thankfully—when you win a deal and you're like, “Wow, I won it.” You're faced with uncertainty, and I think the psychology of that latter moment is very telling.

Have you ever felt that the Andreessen brand holds you back in any way? Like maybe with a—

No. Not one.

It's a massive tailwind, in all ways. There's never been one where there's been a political question.

Not at all. No. And Marc and Ben are so special and authentic. And look, what Ben has said many times is that they feel like it's their responsibility to extend the surface area of the entire industry, and I see them do that every single day.

No, there’s never been even a moment at which it held me back. In fact, it’s been just the opposite.

15. Quick-Fire Round

Dude, I could talk to you all day. I do want to do a quick-fire round with you. Are you ready for this?

Okay.

What’s the most memorable first founder meeting you had? It doesn’t have to be the best founder—just the most memorable first founder meeting you had. And why?

It’s probably the guys at Krea, just because they’d been so mysterious. We’d been unable to get ahold of them for 9 months. They’d been making all this noise on X with their creative tools and their models, and there was so much anticipation around meeting them.

It was Thanksgiving week, so we all flew in. Marc was there. Just seeing these 2 guys walk in as total badasses, with their matching kimonos and their Celsius drinks, and hold the room by being these deep, authentic technologists and product people—it’s just not something that I’d seen before.

There was so much setup to the meeting that it’s one I’ll never forget.

Marc, Ben, D.G.—who’s the best investor?

They’re all extraordinary. Let me tell you the strengths. Marc is the guy who can do 2 things: 1, he’ll paint a picture of the future; but 2, he knows everything about everything else outside of technology. He’s read every book, he’s memorized them all, and he’s got these incredible stories. Of course, he invented the consumer internet, so his storytelling ability is extraordinary.

Ben—to me, The Hard Thing About Hard Things was the first honest business book. I always say about business books that the business model of business books is selling business books; it’s not making you better. Most business books are full of shit, and The Hard Thing About Hard Things was the first one that was authentic.

If you’ve read it as a founder, you’re like, “Oh my God, somebody finally sees me.” His stories of wartime and navigating these inflection points, and his ability to contextualize that for whatever you’re going through, are totally unmatched.

The thing about D.G. that’s so special is that a lot of us are founder-investors. We’re learning how to be investors through the lens of being a founder. D.G. is a pure and highly seasoned investor. He has this pure-play investor clarity that I tend to learn a ton from.

To me, he’s so interesting at the growth stage in the same way that Dixon is interesting at the early stage. So much of our best thinking is Dixon and also D.G.

What’s been the hardest decision that you’ve had to make in the last 2 to 3 years?

To me, a big decision was coming into investing and not being a hands-on builder anymore. I wasn’t sure, because I’d had some incredible investors. I’ve had some investors who just weren’t the best, sometimes through no fault of their own. Sometimes they’re early in their careers, and sometimes they just were disengaged in a way that I never wanted to be.

I was uncertain about whether I wanted to move into investing. I remember sitting down with Ben. Who was I to ask Ben questions? But I thought, “I guess I’m not sure, so let me just be direct with Ben.” I said, “Ben, how do you prevent bad behavior? How do you prevent investor bad behavior? How do you prevent the sort of high anxiety? How do you prevent the person who’s disengaged?”

He said, “In the near term, we don’t measure you based on returns. We measure you by going and talking to every one of your founders every 2 years, doing a 360 on you. If your founders say you’re telling them the truth, you’re showing up, you’re doing the work, and you’re being responsive, regardless of how those companies are performing, then you’re doing a great job. If your founders say anything other than that, regardless of how the companies are performing, you’re looking for a job elsewhere.”

By the way, we do these GP 360s every 2 years. It’s always a little terrifying, but the incentives are all structured in the right way. That moment, in the answer to that question, was when I knew, “Hey, this is not a VC like every other VC I’ve seen out there. This is a company.”

16. The a16z Playbook: How to win 100% of the deals you chase

I remember someone from Andreessen telling me—and I can’t actually remember who it is, so I’m not deliberately being coy. It might have been Brian, it might have been DG, it might have been Alex. I really can’t remember—but they said that at Andreessen, it’s totally unacceptable to lose a deal.

Yeah.

But it’s very acceptable not to have seen a company and for it to be great. Is that fair? To not have seen it? A random company does very well—we never met them, we never had the chance to meet them.

I don’t think so. I don’t think we’re allowed to believe in luck at Andreessen. We have to see 100% of the deals in our domain. Now, look, I think it’s acceptable to make a decision based on the information you have and have the decision be wrong. You invest in a company, it doesn’t always work, and that’s okay. That’s the business.

But the expectation is that we see 100% of the deals in our sector and that we win 100% of the deals that we go after.

That was very clear. No, I love it, dude. Okay, good. Right, that ends that conversation.

I mean, hey, look, some things are ambiguous, but that part is not.

No, no. Good. I hate nuance. “It depends” is the worst answer. You can invest in 1 seed firm. Which seed firm do you invest in?

All right. I think likely Ramtin is pretty special at what he does, actually, at Abstract.

I agree. Why? Why?

He’s a cold-blooded capitalist, which is awesome. He just has great instincts, and I think the seed stage is the hardest stage at which to invest because it’s easy to make 1 great seed investment, but it’s hard to have a system for doing great seed investing.

Even when the people are amazing, there just isn’t anything there yet. In true seed, there’s no product and there’s no go-to-market yet. Now, post-product, pre-traction, that gets easier. Post-product, post-some-traction, that gets a lot easier, and I call that a Series A.

But at true seed, it’s just hard to be right a lot, and he has consistently been right a lot. When I look at a seed manager I respect, I don’t know exactly what his witchcraft is, but it’s working. He’s right a lot.

What have you changed your mind on most in the last 12 months?

I think the thing that surprised me about this product cycle is that I was building my first company in the mobile product cycle. In the mobile product cycle, the anointed winners in 2008 and 2009 were not the eventual winners. We had this cycle where you had the Friendster, and then 2 or 3 years later you had the Facebooks.

In this product cycle, what’s actually interesting is that a bunch of the early leaders from 2023 and 2024 have maintained their lead. We talked about Harvey—that’s a really impressive company. Gamma is a really impressive company. Sierra is a really impressive company. The companies that were early have so far continued to be dominant, and that’s something that I’ve changed my mind on.

I think in 2026 we’re going to see a whole new set of categories. Harry, can I share my view on where we are in the market? Late 2022—November 2022—was ChatGPT. In 2023, a lot of the obviously good ideas—and that’s not to denigrate them; they were obviously good—were started. At the end of 2024, reasoning models started working, so even the ideas that were obviously good but not working suddenly started working with the advent of o1 and DeepSeek. In 2025, those companies scaled.

Now we’re starting to see, for existing markets—which are customer support and the evolution of that, chat, creative tools, and code—we have these early leaders. Those markets are somewhat established. It’s going to be very hard to be another customer-support or coding tool today.

Conversely, I think we’re going to see a set of AI-native categories emerge in 2026. Knowing what we all know now, what company would you build? That is the operative question. Open Claw and Moltbook are just the beginning of that. Those are ideas that were inconceivable 2 years ago.

The thing that I learned over the last 18 months is, “Hey, maybe the early leaders will just be the leaders.” Over the next 12 months, I’m going to pay a lot of attention to who the early leaders are in the new native categories.

How significant is Moltbook? Everyone’s very excited by it. You’re a lot more product-centric than me. How significant is this?

It’s just so damn cool—to talk about it, to observe it. It’s shallow, and likely Balaji called it “robot dogs barking at each other,” and I think there’s an element of truth to that, right? Any humanity they have is just the sparks of the humanity that they’ve taken from the context of their owners.

What is very interesting, though, is the idea that we can have these digital twins—these echoes of ourselves—going and interacting with other people. We were talking about dating downstairs, right, and how dating apps are a mess and probably not durable in their model.

You could imagine a world in which I train a little digital twin of myself. I’m married, but if I was not, I could train a little digital twin of myself, and other people would do the same. They would go have pseudo-dates and then come back and matchmake us and say, “Hey, we had this virtual date, and it went kind of well. Maybe you guys should hang out in person.”

So now we're able to replicate and scale ourselves in a way that was totally science fiction 5 years ago, even 1 year ago. I don't know if you've seen Match.com today, but its stock price is down a huge amount because someone basically did this UGC.

Oh yeah, the Hinge thing.

That's tough.

Yeah, so I think people are looking at the point and saying that we're overhyping it, but they're not looking at the slope, which is being underhyped. I think that's correct, right? Moltbook as an individual data point is probably overhyped right now, but what it points at directionally is underhyped.

That is Tanya's final one. What excites you most? What do you like most? I like optimism. What are you most optimistic for and excited about?

Oh my God, dude. I mean, where do I start? Robots, pet robots. So, for me, it's specifically actually medical. I think we'll have amazing breakthroughs in treatments for multiple sclerosis, which my mom has, which has previously been, like, “Oh, bad luck.”

Yeah.

I think we'll have real breakthroughs there, which is super exciting for me.

Yeah. Okay, okay. Well, let me tell you something personal to myself, which is that I'm a longtime Transcendental Meditation person. I've been meditating since I was a little kid—25, 30 years. It brings me this peace and joy that maybe you see a little bit of in my personality.

I think the idea that everybody could have a little slice of that peace and joy is something that is now becoming more and more possible, because I think that with the technology we have, it's going to take away a lot of the rote parts of life. It's going to give people access to more of these types of relationships that they find so fulfilling.

So, I think just the NPS of the human experience, for lack of a better phrase, is on the way up. I love that for my fellow person. That's what I'm excited about.

Dude, it's such a pleasure to do this in person. Thank you so much for sitting down with me. I really enjoyed it.

Sorry, I'm a little loopy. I can't tell if it's 3:00 a.m. or 3:00 p.m.