The Best Time in History to Start a Company | Patrick Collison & Amjad Masad

The rate of new business creation on Stripe in March, it's up almost 2x year over year. It was one of the early examples of AI companies that went from zero to like 10 million ARR in like a few months, and now it's like a half a billion dollar company. We're going to see more of those billion-dollar kind of single two-person businesses.

Yeah, I just lunched with an economist. This guy Nick Bloom who wrote this very interesting paper actually about how innovation and productivity growth and so forth at least on a per person basis seem to be declining and there was this kind of big stagnation that he was trying to explain. Anyway, we're having lunch and we were talking about how that no longer seems to be true.

For the first time in like decades.

Yeah, yeah, yeah, before AI, before any of that stuff.

And so, it's a bigger spike than COVID and the pandemic and the lockdowns.

And, in terms of, you also had a stat around the monetization of startups. So, that's happening a lot quicker as well, right?

Yes. Yeah, I don't know the exact numbers offhand, but I mean what we've looked at is comparing SaaS companies to kind of the SaaS boom era and marketplaces from the marketplace boom era to AI companies of this era, the time to reach these revenue milestones like a million dollar, 10 million dollars, 100 million dollars even of recurring revenue, it's down by roughly a factor of two.

Yeah. And how much of this is Silicon Valley versus how much of it is decentralizing and there's more entrepreneurship in other places?

Right. Now, it's a little bit difficult to interpret that, right? Because many of those companies didn't start in Silicon Valley, and so maybe the story is it's easier than ever to start somewhere else, and then if you start getting meaningful traction, maybe you then move to Silicon Valley.

Mhm. Yeah, I mean, one of the things anecdotally from the entrepreneurs that I talked to that started their businesses on Replit, the cool thing is that there are a lot of businesses that, you know, Silicon Valley has certain bias, like where a lot of the people that start companies are programmers. And there are large parts of the economy that software has not even reached. Like, we have this entrepreneur in Europe that built software for ice skating rinks. And it's on its way to becoming a multi-million dollar business. He's on his own. I don't think anyone in Silicon Valley has ever thought about ice skating rinks.

Right for sure. We obviously see a lot of Silicon Valley style companies a lot of YC startups things like that. But the more exciting thing is domain experts that have a certain insight in their field or something they're working in that they can suddenly apply software and technology and AI to. We had a teacher during COVID started thinking about education what he enjoyed as a teacher doing. He's been seeing a lot of teachers quitting their jobs and ChatGPT had just come out was like okay how can we use AI to make the teacher's job easier. So he built this product which he called Magic School that allows teachers to create assignments, grade assignments very quickly and all of that. And because he knows how the education system works, you know, a lot of the dogma in Silicon Valley tends to be like don't build a tech businesses because they're terrible. It was one of the early examples of AI companies that went from zero to like 10 million ARR in like a few months and now it's like a half a billion dollar company.

Yes. Yeah, yeah. Well, that's incredible. Yeah, it's awesome. It's really awesome. And it feels like otherwise it wouldn't have been built. Like, you know, but not for what happened with AI, what we built with Replit. And I think that's the thing that gets me excited is, you know, everyone has ideas, everyone goes through life and they see problems in their respective domains but they didn't have the tools to be able to solve them. So what do you think what role does Replit play in the entrepreneurial ecosystem or the business ecosystem in two or three years?

Did you see the New York Times article about like Med V, the company that was, you know, two-person on track to a billion ARR? They built a lot of it on Replit, and I got in touch with Matthew and talked to him about how they used Replit and he... It's a very interesting guy, like he's built a business before, and he had a lot of employees, and he was really frustrated with it. He was like, 'I just don't like managing people. I'd rather manage agents and AIs.' And so, he builds a lot. The first thing I noticed about him is like from idea to like prompt on Replit, something is up in like 5 minutes. It's like the fastest person to just spin up projects and ideas and things like that. There's a lot of automations that he was building on the back end. He obviously uses Zapier and a lot of other things. But it's just got me thinking how much of after you start getting customers, how much chores there is in running the business. So, I'm starting to think about a layer on top of your Replit projects. You're already running your business, your traffic is there, your users are there, your customers are there. You know, can you launch a marketing campaign really quickly from there? Can you automate your inbound sales process? I suspect there's a lot of value to that vertical integration.

Yeah, my mental model for Replit is like a Nintendo. Like you can go on you can buy it like I bought a Steam Deck and like you install the games, you know, half of them don't work, you still have to, you know, configure them and always tinker with them and somehow they break sometimes randomly. And then but you buy a Nintendo and yes, you have a lot of other games on the Nintendo, but you're also, you know, play Zelda, it's like the best experience ever because they control everything from the pixels down to the...

Yeah, that's right. Yeah, that's the exciting part and I think we're going to see more of those like billion-dollar kind of single two-person businesses.

Right. I mean, I think it gets pretty philosophical pretty quickly. Like, what is special about humans? My view is that at least the current era of LLMs and it doesn't look like it looks like LLMs will just continue to scale with almost the same properties, which is, you know, they're just like a still a function of their data, right? That means they're great machines for doing what already existingly works, but not necessarily what's on the edge of what's possible, what's on the edge of culture. That's why they're not very good at like naming or marketing, things like that. And I think the role of the human is to like have a sense of what other people want, what's there in the, you know, there's an entrepreneur that built like a looksmaxing app on Replit. Like, ChatGPT would not even know what looksmaxing app would be like, and that's like a moment in culture. And so a lot of it is about just finding these opportunities and executing on them.

Close system. Yeah, exactly.

Yeah, I'm really curious about this. And it feels like we'll probably learn the answer to that in the next couple of years. You know, we're sort of on the cusp now of these Erdős problems being solved and so on. So, we're, you know, the models of decent math are really getting to the frontier and I think it's really a very interesting question. Do they overshoot and, you know, in a couple of years they're just way beyond the frontier and to your point they've invented these completely new fields of math that humans have not even conceived or are they somehow the world's best ever contest mathematicians that can kind of operate more effectively than anything else in the framework, but somehow can't frame break and, yeah, I'm excited to know the answer.

Yeah. Um, hm. I have a lot of affinity for the vertical SaaS ideas. That is to say, taking some space that has not been massively amplified and transformed by software and really bring it to bear. Or in many domains, there's a really janky, annoying, cludgy 1990s thing that's kind of dominant, but it has a terrible mobile app and zero AI support and terrible collaboration features and, you know, whatever. And so just going and building the good version for that particular domain. So I think there's still a ton of opportunities there. And then I think there's something, I mean I think this can be a little bit risky, but I think a pretty interesting pattern is what's popular among young people but low status. Obviously crypto was that back, you know, 10 or so years ago. I mean maybe it's still that to some extent today, but certainly in the past it was and such that it's overlooked by the incumbents. And I mean even startups as a market were that 15 years ago, right? Like Stripe sold to startups. And we had no competition because nobody else thought that was an important market because we, you know, didn't have all the success cases that we now do. So I think those would be, I think number one, just try anything. Number two, just look for other domains where software hasn't really been brought to bear in an effective way. And then number three, yeah, what is popular among young people but low status.

Goodness. I think...

Want to continue going down that path of questions because I want to make sure like people in the audience are, because I think there's a natural tendency to always like come up with, you know, counterexamples or things to say why not to do something. So, you know, there's also this trend of saying that the, you know, foundation lab companies are going to be able to capture big part of the at least new business creation, but perhaps the economy. It's like there's no reason to compete with ChatGPT because or Claude because they're just going to roll over us.

That's a very Marxist argument to this. Yeah, I just yeah.

Right. And I think one thing about LLMs, and who knows, it might change, but one thing about LLMs it just feels like a very low-level technology in the same way that computers, the interface is sort of like Linux, you know, text in, text out, and very commodifiable interface. And the history of LM progress has been, you know, GPT-2 was something only OpenAI could do. I remember 2019, everyone in this town was freaking out about GPT-2, no one else in the world knew what it is. Now you can train it on your phone. You can train a GPT-2-like model on your phone. And so there's like a Moore's Law-type progress that's happening in AI, and it feels like the technology is going to be permeating into everything. There's going to be edge inference. There's going to be a lot of open-source models. Perhaps it doesn't feel like it right now, but it's hard for me to imagine a world where it's only a few companies that are controlling these, you know, AGI-like models that can do everything, and they're also building all the interfaces on top of them. And by the way, I think also that the way people interact with technology, you know, I think part of the argument that, you know, the models will just eat everything is that it assumes that everything is, you know, the best way to interact with AI is via just like prompt, you know, just put in a prompt and something happens. But I think people would still want interfaces. I think interfaces are very sort of a human thing. They want creative interfaces to be able to interact with software, and they would... There's probably new types of platforms that are going to come up like AR, VR, and I think computing will change a lot, but will change in the same way it has always changed.

Okay, so I was going to ask you this. What do you, I mean, to what degree if at all do you think in 5 years or 10 years the shape of moats in software will have changed?

Yeah, and it's basically it's like there's seven moats. And you go back in history or you try to predict forward in history and it's, you can map these seven powers on whatever business you study and typically businesses that are valuable continue to be valuable and capture a lot of value and maintain their defensive position in the market tend to have one of those moats. Does AI change any of those moats?

I think the meta point here would be that through every decade in Silicon Valley there have been significant overhanging concerns about entrepreneurship. Like in the '80s was Japan is going to win at everything and then in the '90s Microsoft is going to dominate everything and then in the 2000s was after the dot-com crash and so maybe just like the internet is useless and there's no point trying anything. And then there was Google and I came here for Stripe in 2010 and like all through the 2010 it was yeah some combination of Google and Facebook and concerns there but then also people were always and you remember this always talking about the funding drying up and, you know, the bubble had to crash and, you know, it always felt they were kind of somehow on the precipice of disaster. Then it's COVID and now it's these AI labs and obviously through the entirety of that almost now 50 year period it's actually been empirically a really good time to start a company and that like many amazing companies were started and so like not only do the current short-term trend lines seem incredibly favorable and encouraging but I think if you just zoom out and take a slightly longer snapshot of the history here it's, you know, for more than a generation betting on the side of entrepreneurship has been a pretty durable bet.

More consumer surplus.

I'm very bullish on yoga and yoga instructors through the singularity.

Yes, absolutely. Awesome. Awesome. Thank you so much.