The Exchanges

Q Because you're surrounded by people who've done so much, you feel like I haven't done enough?

A Yeah, well, I've always, I've never felt like I've done enough, so I remember I was, like, 16 and I was having, like, an existential crisis of, like, I'm 16. I haven't done anything yet. What's going on? I've ruined my life, you know, and so I think it's a personality trait partially, and then partially as you see these really cool things that people are doing, and you're like, I want to do more, right? Like Patrick from Stripe started ARC, a biology institute, like one of the first new major biology institutions in decades that's doing really interesting things, right? Or, um, Brian Armstrong started a new anti-aging company, or, you know, so you, you see what people are doing. You see what Elon Musk is doing, you know, and so it's very inspiring To try and build more for society, do more for people, think big. You know, I think it's important to, so I don't, I, you know, I don't feel like I've, I can rest and go and whatever.

AI assessment note: “I think it's a personality trait partially, and then partially as you see these”

Q And if more people are building the next Minecraft or the next HubSpot or the next all of these things, are they all going to end up being much smaller players because there is so much competition now and so many people tinkering and creating? Where do you see it going?

A I think the biggest things in the world are only getting bigger, right? If you, and there's more aggregation than less, and there's a long tail of stuff that can get much bigger than it used to be able to get. And part of that is, We've moved to global liquidity on the internet where we went from, you know, ten million users to a hundred million users to billions and billions of users, and so suddenly you have this global market that can really accelerate adoption, right? OpenAI, Anthropic, and others are the fastest companies to ten billion dollars in revenue basically ever, right? And the market caps, if you go back 15 years, the biggest market cap in the world, I think, was Exxon, and it was like four hundred billion. Right? And now we have market caps 10 times that size, which nobody ever expected. There's eight companies now worth over a trillion dollars each, which back then nobody would have thought was even possible. I remember talking to them like, oh yeah, maybe something will someday be a five billion dollar company. What a giant outcome, right? Um, but the biggest things have gotten way bigger, right? And so Microsoft is way bigger than anybody thought. Google is way bigger than anybody thought. It would be Meta, et cetera, Nvidia. Um, so if anything, value is aggregated up much more than anybody expected. And the question is, does that continue or not? And I don't …

AI assessment note: “I think the biggest things in the world are only getting bigger”

Q that's off the shelf dead? Did we go from software that's literally off the shelf at Staples on a CD that you bring home to software that's on the web that you download to your computer to software that just exists on the web to now no longer software that exists on the web software in your imagination and no one's going to pay for this. What do you think?

A I think it's going to persist or some subset of software persist. Obviously some will get eaten away by AI, but if something works and it's already been Um, battle tested for security and a bunch of other stuff, and it integrates with a bunch of things. Um, if it's cheap enough, you'll just keep using it. Like, why even spend time on it or worry about it, right? So part of it also comes down to, like, what's your relative cost of something, and how much do they charge you, and how, how onerous is it to replace it, and if you replace it, are there other dependencies, and eventually, you know, your coding agent will be able to deal with some of those dependencies, but is that the highest value thing for your coding agency to do? This is why internal tools often aren't, don't get, have giant teams at companies, right? It's because they want to deploy those engineers against The bigger product development, right? The bigger fish to fry. And then they realize that, oh, we really need internal tools and we need internal tooling team. And sometimes those are amazing people and sometimes the people aren't as strong because they're really strong people want to work on product, right? The coding agent thing may be similar in that you may only want to deploy them against the things that are highest ROI. You're going to have some limited budget for what you're doing.

AI assessment note: “I think it's going to persist or some subset of software persist.”

Q through a tough divorce, so he's got a divorce AI. But if he were going to productize this email marketing customized solution, maybe people wouldn't sign up to that. Maybe they would say, I like that. Hey, Alex, build the same thing for me. And then Alex builds it for five businesses. He keeps adjusting and he keeps improving it based on what he's learning. And is that the future?

A Well, it sounds like it's just repeated, building repeatable software then. I don't know, what's the difference? Where's the line? Is it five customers versus a thousand? Is it, you know, why doesn't he just start selling that to everyone? So, you know, it just sounds like a different size customer. Now, if you look at certain types of industries, government, certain really big institutions, et cetera, they always did this. They're still doing it, right? The, um, I don't know if it was the FBI or CIA or somebody spent like eighty million dollars to build an ATS. They hired a third party consulting firm. I remember here with Accenture, Deloitte, someone built them an ATS for like, eighty million dollars. They could have just bought Greenhouse or in a more recent era Ashby or something.

AI assessment note: “Well, it sounds like it's just repeated, building repeatable software then.”

Q Is it just introducing, like, ideas? Are you building software for them? Are you helping them buy software? How does it work?

A It's, uh, building stuff. Yeah, it's basically asking what are, um, things that would be very high ROI to implement? What would, what would create good returns for these customers? And what can be done in an expedient manner based on what we've built to date? So that that way we have some, you know, a set of modules that we're using that we've already built. There's a platform that we've already built. And so how do we rapidly accelerate something so that We can show very fast success, and then once we show success, we can add the next thing within a vertical and build against that, and it may take a little bit longer, but now they trust us, they know us, you know, we may have access to data or other things that can be used to sort of implement these other things with the idea that that basic fundamental module tool can be customized and cross sold to other customers in the same vertical over time.

AI assessment note: “It's, uh, building stuff. Yeah, it's basically asking what are, um, things”

Q like vibe coding this thing, or somehow he's coding it. You're, you're saying There will be a software provider that does it. It'll be the new MailChimp. If you're, if you have that idea, don't give up on it and say it's going to be eaten away by vibe coding software and people will create it themselves. Create that software. There's a future in that. Am I understanding you right?

A Yeah. I mean, he may create the next MailChimp, right? I don't know. So that, that's the question is, um, you know, where does, where does the innovation come from and where does scale come from? And in general, what you find, I mean, this happens all the time, right? Where, say you do have an internal tools team building something for you. This, this happened with a company that I helped out with quite early called Braintrust, where, um, you know, they basically provide like an eval suite for enterprise amongst a bunch of other stuff, right? They have sort of a series of products now, but what they kind of started with was, um, can we help make eval really easy as you adopt the LLM so that you can change the model and you can see what are, how do those changes propagate in terms of performance relative to some test stuff that you have? And, um, and, you know, a lot of companies are using them now, uh, in terms of sort of the leading companies that have adopted AI. And I think that, um, that was a good example where I helped them with early customer calls, and we call a customer, and, um, those customers would often say things like, um, oh yeah, I already have an internal team working on this, we don't need it. And then they call back three months later and they'd say, actually, we really need this. Our internal team, the resources should go to something else. This is important…

AI assessment note: “Yeah. I mean, he may create the next MailChimp, right?”

Q What do you see consumer that's exciting? It doesn't seem to me like you, I don't know, I felt like maybe at the beginning that you weren't as excited about, and I don't see you in consumer tech much. What do you see in consumer that's exciting?

A Yeah, um, I used to do consumer tech. I was, you know, I invested in Airbnb and Instacart and a bunch of stuff quite early, and then I worked at Twitter for a while. I sold the company to them. Um, but then I think there was a gap where there wasn't that much, and, um, you know, one could, I think a lot of the kind of prosumery things are a mix of consumer and Professional use cases. So early mid journey was both of those things, right? You do it for fun and for art, but you also use it professionally. Perplexity is that way. Open AI is that way. Um, you know, I think, I think there's a lot of blurring. Um, you know, my hope is that we see more and more companies tackling big problems. I mean, you know, there's stuff around email, there's stuff around productivity, there's stuff around personal agents, there's, you know, there's a lot to do. Now the question is, does a big company do it or does a startup do it? And that's the hard thing with consumer right now, because the difference between now and 15 years ago is there's a handful of companies with such large distribution that they're the natural homes to do new things. Because even if they're five years late, they can just cross sell it, right? And that's often what happens, right? In the early days of a startup, you're competing with other startups. In the later days of a startup, you're competing with incumbents. And so th…

AI assessment note: “a lot of the kind of prosumery things are a mix of consumer”

Q I still have not found an AI company that's, never mind a one person, one billion dollar company to interview. I haven't even seen it at like one person, a hundred million dollars, maybe one person, fifty million dollar company. I got one. He won't come on here and do an interview. Are you seeing these and maybe I'm missing them?

A I'm not saying a lot of them, but that's also, um, one could argue that in Silicon Valley, which is still a lot of the activity or in tech or whatever you want to call it, Silicon Valley is a place, but I was thinking of it as a concept, um, uh, People tend to get on the venture train too much or on the let's raise money train, right? And so, definitionally, they raise money, they hire people, they don't really bootstrap much, um, and one could argue outside of Silicon Valley, people raise too little money. There's too little access to capital, and so often the ideas end up smaller than they should be, because people try to get profitable right away, they aren't very aggressive, And so to some extent, one could argue Silicon Valley is our being builders into environments where they're fueled with capital, and hopefully they get much bigger than they would in other domains or areas or regions, but the flip side also happens where companies will blow or grow faster than they should in many cases or could, and, or they'll misprice things because they don't really think about their pricing power, right? When, you know, if you charge a lot up front, I remember talking to, um, I can't believe it was Olivier from Datadog or something like that. They basically said, we always charge a lot for our products, because we were worried about profitability. And so they ended up as these cash …

AI assessment note: “I'm not saying a lot of them, but that's also”

Q talking to you before we got started about how Paul Graham said, look to people who are creating toys, that's the future, and we look at people who created these little things that almost seemed like they were features on somebody else's thing, and then they ended up becoming big companies. And I'm wondering where the toys are today that will become that big. What do you think about that?

A I think there's two ways to interpret what he said. One way is to say, look at the people who are tinkering with interesting things, and those things often become really big, and that was the early microcomputer revolution, the things that turned into PCs was just hobbyists in the seventies. Um, you could argue that GPT one and two kind of felt toy-like, even though there were these clear scaling laws behind it, right? But they couldn't do that much, but it's super interesting, right? Um, so there's no sorts of toys where people are working at the frontier. There's a second way to interpret work on toys, which is basically, if you look at the consumer internet wave, those things truly felt like toys. You know, Twitter was like, text your friends. Like, literally, you could only text. Or, you know, Airbnb was like, crash on somebody's couch and pay money for it. You know, and originally it was literally like, um, it was called the air bed and breakfast, because it was literally, you know, mattresses that the founders wouldn't place, so they could have people stay with them during design conferences and monetize it, right? So I think that era of consumer internet has moved on, and you don't show up with, to, to, when you're selling, like, a SaaS product with something that's a toy, right? You show up with something that's fully baked, or that works well, or that solves some use c…

AI assessment note: “the toy phase, at least in the context of, Products that most people are building”

Q Well, I see. Maybe the definition is different. What do you think? Yeah.

A Even, um, You know, back to Twitter, it was like, here's a really sloppy looking site, and here's a short code, and you can, you know, the 140 character limit is because that's all that could fit into an SMS. It wasn't because they were thinking of, like, gravity or something. Um, and then it kind of stuck as a feature for a really long time, much longer than it should have, right, because it was an accident of just SMSs having character limits. Um, so I think there's almost like this, um, I think it depends on then what you mean by toy. If it's like, Hey, I'm going to build something really quick and simple and dirty. And honestly, that was all the vibe coding stuff. Like there used to be these really bad demos on GPT three of, Hey, look, I'm going to type in some stuff and make a really shitty UI for me. And now we have lovable and replet and Figma and all these companies like providing by coding. And so, you know, that, that felt toy like back then, but everybody I knew was like, Oh wow. If these models got much better, this is going to be amazing.

AI assessment note: “I think it depends on then what you mean by toy.”

Q Because you're somebody who's such a big thinker, right? Rocket ships. How do you get people to live longer? But at the same time, I'm curious about how you see the world. And I want to see that understanding by seeing this little example. If you had all this, what, what would you start? What would you build? How would you think about the fresh new idea to get into?

A Yeah, it's, it's hard to answer that, because there's a lot that I think should be built right now, and so, and, and there's a wide range of stuff. I mean, it's everything from new foundation models for materials, like what Periodic is doing for physics, for a variety of areas, so I think there's a whole, like, set of models that nobody's built yet, or that I should say people are building, but it's very early. That I think could be incredibly impactful and interesting and just like cool, you know, like it'd be super interesting to reinterpret physics through the lens of like a, you know, a model or a modern sort of architecture model. Um, so I think stuff like that's super interesting. I think if you go up a layer or two, there's so many things to do on the application side. Again, I think it was a real dearth of consumer products and there's a ton to build there and all sorts of interesting ways. Actually, um, me and somebody on my team ran this program at Stanford. Probably about two years ago, maybe a little bit more now, three years ago, where I think GPT-IV had just come out, and we said, um, we'll give you free, uh, compute. We don't want anything in exchange. We just want you to run experiments and build cool consumer stuff, and we had about a dozen teams from Stanford go through it. It was all undergrads and grad students and CS and AI, and they, and we'd have, like, a…

AI assessment note: “everything from new foundation models for materials, like what Periodic is doing for physics”

Q Okay. And I asked you before, can we please announce some kind of news? And you said, no, Andrew, I got nothing for you.

A Well, that's not anything that interesting. I think, um, you know, there's some things that we're moving towards, like our, our hope is to announce our first sort of statue or monument soon. We've been working on a project to do that for like societal beauty, like art and society at large scale. And so, um, there's some stuff getting pretty close there. There's a few things that we funded that, um, haven't announced their rounds yet. So, You know, there'll be things like that, so that, you know, there's a bunch of stuff going on, but I don't think there's anything I can really talk about yet, but I'm happy to chat again if you'll ever have me, and you know.

AI assessment note: “our hope is to announce our first sort of statue or monument soon.”