The Harsh Truth About Building a Startup in the AI Era | Anish Acharya

And every founder has that feeling like, 'Oh my god, I'm too late.' And now it doesn't matter what we build because no one will fund it. But if you're able to put the blinders on and just turn that noise down for a moment, what you'll see is actually we're in the best time to build a startup we've ever been in. It is objectively the fastest developing product cycle we've ever seen. You can do these incredible, extraordinary things that would have sounded like science fiction five years ago. But there's just so many places that you can get stuck. And I'll tell you about some of the traps. One of the traps says...

The whole trick of being a founder, much of the trick is an exercise in psychology. The one mistake I'd encourage folks not to make is underestimating how big these markets are. Like these are not markets, they're industries. If you look at what happened in 2011, 2012, we suddenly had this explosion of Lyft, Uber, WhatsApp, Airbnb. We're in that moment again. Millions and millions of users are downloading consumer AI products organically. Consumers are incredibly excited about AI despite what you read in the newspaper. And then the other cool thing is for the first time, you've got consumption revenue for consumers. Typically, in the past, a consumer could subscribe to a product, but you're never going to pay Spotify $100 in a given month. Whereas now, if you look at Gemini Ultra, Google Ultra, that's 250 a month. People are paying it. The magic right now is that you can lead with product instead of marketing dollars. Right now is the moment to be at the edge. And in terms of the exact trends we're seeing, the most important trend, the one I'm personally most excited about is...

We're working hard. We try to be everywhere, you know, and when there's more opportunity, that means, you know, Ben says there's no such thing as luck. We're not allowed to be lucky. We've just got to cover everything. So, when there's more things to cover, we've just got to put in more cycles, you know.

Yeah, you know, it's funny, Ellie, because Andreessen is so well known, but the kind of core thesis of the firm, I think, is less well known. I didn't know it when I joined in 2019. And the core thesis of the firm, Marc and Ben have written a lovely blog post about this that, you know, we should refer your readers to, but the core thesis has always been that founders make the best long-term CEOs and leaders of their companies, right? And that sounds like a non-controversial statement, but 10 years ago, when the firm started 15 years ago, it was actually controversial. The common practice at the time was that venture firms would bring in a so-called professional, experienced CEO. Now, what is the difference between a founder and a CEO? It tends to be two things, right? Knowledge and networks. So, the whole structure of the firm is meant to help founders be those CEOs of their companies forever by building networks and knowledge around them. So, that's like the core thesis. That's a good sort of foundational building block. And then in terms of the way the firm is composed, it is a big firm, but everybody is a specialist. So the expectation is that we're all very, very deep and kind of obsessed and technical and operational and well-networked in our areas. And because so many interesting things in software cut across many areas, we're able to kind of come together and bring a better proposition to the founder than you would get from a smaller team of pure generalists. So that's how we work.

The team size is large, but the way to think about the team, so I think the overall firm is maybe 500 people. But the way to think about this is, you know, the original sort of Andreessen, which is all the GPs in one room and all the sort of operating teams around them. We've sort of replicated that. So today, our apps fund that I invest out of, you know, looks a lot like Andreessen Horowitz period 10 years ago, the infra fund, the American Dynamism fund, etc. So we've really taken the structure of a small number of GPs, operating specialists and other folks and replicated it because software is just now so much more horizontal than it was 10 or 15 years ago. So that's the kind of ethos. You know, the energy that you get in working with us is no different than 10 years ago. You just have more specialization.

Yeah, Marc said this recently and I think it's the right framing, which is the thing you want from your venture firm is power. And the reason for that is, you know, when you're a tiny, you know, baby bunny rabbit startup, you know, you don't necessarily have the types of power that you need, you're not necessarily able to phone the Fortune 500, the Fortune 2000. You don't have brand power and presence. So, what you really want to do is take the platform that your investor gives you and build on their power. And then over the course of time, if things go well, you of course become more powerful and more significant and more well known than your venture investor. So, distribution is a part of that. But I think broadly what you want to do is borrow the sort of credibility and the presence that your investor has, and we really try to create the platform that founders can jump off of.

Well, you guys have done it really well. You know, when you have somebody like a Maddy on your pod and then you've got a lesser-known founder with a smaller scale company, you know, it's sort of the collective presence that you and Maddy and the other guests that you've had really prop up that founder that may be earlier stage. So, it exactly mirrors what we're doing. And, you know, it really benefits the ecosystem as a whole. And by the way, I hope more firms do do it. Like I think we're going to have a lot more software and we need a lot more founder stories to be told.

I'd love to. Yes. Yes.

I know. I mean, I think it is objectively the fastest sort of developing product cycle we've ever seen. There's a lot of reasons for that. You know, one of it is just the size of tech is so much larger than it ever was before. You know, I think for a long time tech was the dog chasing the car and now we are the car or something. We've really fulfilled our ambitions. We're no longer these kind of, you know, weirdo hippie kids out in San Francisco or Palo Alto or whatever else. So like one, the scale of tech is a lot bigger. Also I think the nature of this product cycle is different. Like if you think back to the mobile product cycle, it sort of was all downstream of Apple. So Apple did the iPhone in 2008. They did the App Store in 2009. This was a surprising fact to me, but when the App Store came out in 2009, there was only six million iPhones in distribution. Right? So one year post iPhone, your TAM as a mobile app developer was six million customers. Today, if you build into the Apps SDK, which is the new AI app store that ChatGPT is going to be releasing, you've got an audience of 850 million people. So just like the scale is enormous. But the other interesting thing is when we were building an app in the App Store in 2009, 2010, what happened was Apple sort of decided what APIs they would expose and all of the developers would organize around those APIs. Whereas it feels like with large models, it's more of a process of sort of biology than systems design in that the models are being trained, we're discovering what the models can do, and they have all of these emergent properties. So I think for that reason, there's just this much wider diffusion of things that are being built, things that can be built, because no central decider is, you know, upstream of the entire innovation ecosystem. It's really, really crazy.

I mean, I think that the whole trick of being a founder, much of the trick is an exercise in psychology, personal psychology. I think there's just so many places that you can get stuck, you know, and I'll tell you about some of the traps. One of the traps is, 'Oh my god, I'm too late,' you know, and every founder has that feeling like, oh, I should have started my company a year ago, a month ago, you know, and if I had, I'd be where XYZ is. So, I think that's a very common trap. Another common one is, you know, nobody is funding anyone. You know, we're in a winter. Oh my god. Like even if I build something, nobody will pay attention. When I was building my first startup in 2009, I remember the TechCrunch article. Well, the sort of headline was the end of venture capital. And I sent it to my co-founder and I said, 'Ah, it's over.' You know, like venture capital's over. TechCrunch has said so. And you know, it doesn't matter what we build because no one will fund it. Looking at competition is another sort of mind trap where you can look around and feel like, 'Oh my god, look at how much everyone else is raising from all these credible folks.' But if you're able to put the blinders on and just turn that noise down for a moment and look at where we are from first principles, what you'll see is actually we're in kind of the best time to build a startup we've ever been in, that I've ever seen for sure. Which is, let's say you're building a consumer. What is the nature of consumer product today? Consumers are incredibly excited about AI despite what you read in the newspaper. Millions and millions of users are downloading consumer AI products organically. So there's no paid acquisition necessary. With these large models, you can do these incredible, extraordinary things that would have sounded like science fiction five years ago. And then, you know, the model companies are beating each other's brains out to kind of spend more to improve the models that really benefit all the founders and builders. We just saw that with Gemini 3, which looks extraordinary, was released yesterday. So, it just feels like from a first principles perspective, there's never been more opportunity, even though to your point, there's never been more noise.

Good for them.

Well, I'll tell you all, the mistake that I have consistently made, and I feel like I continue to make it in this AI ecosystem, is underestimating how big these markets are. And that's why the way I describe it is like these are not markets, they're industries. Okay. So AI code is not a market with a single winner or even a half dozen winners. It's going to be an industry with, you know, 30, 40, 50 winners. Legal, AI legal is going to be as significant as simply legal. You know, it's like we used to say e-business, but it's really just business. When we say AI legal, we really just mean legal. And I think it's going to be enormous. Yes, there's going to be specialization, but the size of those specialized markets are going to be as big as entire software markets were previously. It's definitely not too late. And I think that the one mistake I'd encourage folks not to make is underestimating the size and significance of the opportunity.

Yeah, and I have to sort of quibble with the framing of AI is going to start replacing jobs because we're out in the front line talking to all of these AI companies that are selling enterprise software. And what we don't hear very often is that fully replaces a job. What we do hear is it fully automates a task, but tasks are not jobs. So, what we've typically seen with these enterprise software plays, even ones that when you squint out from afar and say, 'Oh, wow. The AI is answering the phone and doing a lot of negotiation, like maybe that's going to replace a job.' What we actually see is that it gives the people that are doing the job a lot of leverage. They no longer have to do the sort of administrative, rote parts of the job. They're able to specialize and, you know, perhaps those folks are, you know, spending less time than they previously spent overall working, but we're not seeing role elimination broadly. So I would kind of quibble with the narrative of AI is replacing jobs. I think AI is automating tasks. As for the next thousand days, you know, if you look back at December of 2022, you know, just, you know, D30 after ChatGPT launched, I think there was a very valid concern that if a single company was an entire generation ahead in terms of model quality, then there was all kinds of economic implications of that, including that they could take a lot of the margin from the entire AI ecosystem, that they would sort of decide what everybody would build, what constraints, what opportunities, etc. But it's just not what we've seen over the thousand days. Instead, we see, you know, a half dozen really significant cutting edge models that are all improving neck and neck. We see a bunch of open source efforts, some from China, some from the US, all of them very interesting, all of them pointed in different directions. So, it feels like there's been this diffusion of models, cutting edge models in particular, that mean the kind of dystopian predictions about what would happen if one company controlled all of the AI so far haven't played out. So, I think the next thousand days will be extraordinary in terms of model capability improvement, but I don't know that we're going to suddenly see, you know, some sort of runaway dystopian effects from the models. I think we're just going to see more of what we've already seen.

100%. Like what we've actually seen, and one of our best companies called Happy Robot does this, where they sell voice AI to freight brokers, you may be familiar with them. And what they've actually seen is for a lot of the folks that were previously doing, you know, just low-level customer support work, they're now moved into customer relationship management where, look, like the AI is never going to take a key customer account out for a steak dinner. Somebody's still got to do that. But if people are too busy handling kind of the day-to-day minutiae, they have less time for relationships. So instead, we're seeing, you know, humans get to be more human than ever. And AI takes away a lot of the work that really didn't get leverage from, you know, human skills anyways.

Based here, but Spanish founders. Yeah, they're tremendous.

Sounds excellent. Let's do it. Where should we start?

Yes. Yeah, that's exactly right. So consumer is hypercyclical. So for long, long periods of time, nothing happens and it's a very tough time to kind of invest or build in the consumer ecosystem. And then you have these magical moments where the window opens up and all of a sudden you can build consumer tech and all of a sudden the consumers change, something is in the water, new products are being launched, the sort of boundary of imagination is being pushed. And born from those moments in time are some of the biggest companies in the world. You know, if you look at Mag 7, the majority of those companies are consumer companies or have a significant consumer offering. If you look at what happened post iPhone, as 2011, 2012, we suddenly had this explosion of, you know, Lyft, Uber, WhatsApp, Airbnb. It was just extraordinary what happened. We're in that moment again. And there's three things that really unlock consumer markets. You know, the first is a new technology, which we of course have in AI. The second is new consumer behaviors, which is connected to new technology, but people just making different choices. You know, in the mobile era, I remember people saying, oh, location is too sensitive. No one will ever share their location. And now every Gen Z shares their location with all their friends and their exes, and it's very strange behavior. But you know, the consumer choices and preferences just change very quickly. And then the third is a new distribution channel. If there was a critique of consumer tech at this moment, it's that we don't really have an AI-native distribution channel. I think that's about to change in 2026, and we can talk about that. But we sort of have the ingredients of really magical sort of consumer growth and consumer startup building right now. In terms of the exact trends we're seeing, I think the most important trend, the one I'm personally most excited about is, you know, consumers building software. If you think about the internet, the internet has been the sort of driving force for participation. You know, everything from written content, which transformed into social media, to video content, which transformed from shaky home camera films into what we now see in YouTube. I think I read that YouTube's a $550 billion enterprise. Like, that's pretty crazy. Built off the backs of, you know, MrBeast and, you know, unboxing videos and things that were entirely non-predictable 20 years ago. It feels like the software ecosystem is headed to a similar place. And you know, you asked a question of like, will there be a new sort of social network? Maybe there will be, but I think it'll be more focused on things like software versus content. And we're seeing the beginnings of that now. You know, we announced an investment in a really cool company called Wabby a few weeks ago, W-A-B-B-Y, which is a platform where consumers can create and consume personal software, mini apps. It feels exactly like YouTube in 2006. And it's easy to look at it and say, well, hey, do people want to make software? What kind of software will they make? Can they make really serious software? I think that's the same thing as looking in 2006 and saying, 'Hey, how big and how important can consumer video be?' So, the idea that people will make software for themselves, for their friends, the idea that you've got 20 million people who are programmers today and 6 billion people that use software, like let's change that ratio. You know, why can't there be 500 million or a billion people making software? And it just feels like, you know, Lovable, Replit, Codex, Claude Code, Wabby, like Emergent. I mean, you can Vibe Code, Rork, you can just name the companies. And the amazing thing is they're all working. And the reason they're all working, I think, is that there's just this huge sucking sound of demand from the market that no matter how many new startups cross 100 million of revenue, there's like room for another dozen.

Eugenia.

Eugenia. There you go.

Yeah. Yeah. Well, I think it's two things. I think, you know, top down, business model quality is really high. So you can just, as a consumer founder, you have to ask yourself like, how am I going to spend every day? And I think as a consumer founder from 2014 to 2023, the way you probably spent your day was marketing, you know, and that's just very few of us are people that get joy from consumer product sort of marketing and customer acquisition. Whereas if you look at how you spend your day today as a consumer founder, you're spending it building product. I would argue that there are no marketing problems today for consumer companies, there are only product problems. And if you're not getting distribution, you probably haven't been ambitious enough in product. So, I just think that this is a really fun and magical time. You know, business model quality is high. Consumers are willing to pay. They want to try new things. And then you have so much to work with as a consumer founder. You know, you're able to build products that are emotional and sort of capture and understand emotional experiences. Products like Poke, you know, you're able to do things with new sort of primitives like voice. I mean, voice is the original form of human communication. And yet there's really been zero tech built around it ever. Like what will the future of voice as an interface look like? I don't know. But you can kind of invent it and imagine it today. So it just feels like the amount of fun you're having and like the really exciting consequences of being successful and having that fun are something that we haven't seen since 2010.

Well, yeah. And actually, I think $200 is like, 'Sure, I'll sign up tomorrow' for many people. And those are the price points. If you look at, you know, Gemini Ultra, Google Ultra, that's 250 a month. ChatGPT's top SKU is 200. Grok Heavy is 300. People are paying it, you know. I mean, if you look at the spend on Cursor, I'd love to see what hobbyist spend is on Cursor. I bet it's really significant. I think it's easy to look at their revenue scale and say, 'Hey, it's enterprise.' And I know I'm sure a significant part of it is enterprise, but I'm sure a lot of it is also hobbyist. The other cool thing is for the first time ever, you've got consumption revenue for consumers, right? So, typically in the past, a consumer could subscribe to a product, but you know, you're never going to pay Spotify $100. If you pay 20 a month in subscription, there wasn't a way if you listen to 5x more music for you to pay them $100 in a given month. Whereas now, if you're using a creative tool like Crea, a coding tool like Cursor, you can conceptually pay that company a lot more than the fixed subscription rate. And again, that's kind of a new business model improvement for consumer founders.