Agents Write Most of Our Code | Replit CEO Amjad Masad

Thank you. Thanks for having me.

Yeah, we try to create a one-stop shop platform for making software. So our vision is that anyone of any skill level should be able to make software that is safe, secure, scalable. Like there's no reason at this point for anyone in the world that has an idea not to be able to build it and ship it and monetize it. And so we take every aspect of the software development life cycle and try to make it more fun, easier, more collaborative, safer, secure. Security has been a big thing we've been focused on recently.

Yeah. I think, you know, so I always believed in my heart of hearts that, and I also had evidence that if you make software easier to create, there's enormous economic activity to be unlocked, right? And we are all seeing that right now. I mean, I'm sure a lot of you are examples of that. Supabase is certainly an example of that. And it just seems obvious that if we kept going down our mission, that we'll be able to get to a place to achieve that mission. And, you know, there was like a fair amount of luck that we were riding certain waves of technology. So I think the cloud wave is one, right? When we started, there was like barely any like Docker or Kubernetes, and we had to build all of that from scratch. But then all of the stuff came online, it became a lot easier to do. And then AI, like we started investing in AI back in 2020, and it was horrible. It was like magical, but it was still horrible at the time. Like GPT-3 would like put out like two lines of code and start speaking Chinese. And so we had to do a lot of wrangling. Like actually, I was reflecting, our harness today is 10 times simpler and 10 times less code than our harness in like 2024.

Yeah, I mean, so if you make programming easier to get started with, but the environment is pretty limited, what kind of use cases do you unlock? Maybe a little bit of scripting use case, but the biggest use case is people learning to code. And in fact, that's why I built Replit. I was initially in college, kind of trying out a lot of different programming languages, doing some homework, or like, you know, I need to like be able to code in the browser. I shouldn't have to install like freaking IntelliJ for the thousandth time and download gigabytes of software on my computer. So I want to be able to like get a browser-based Replit, and surprising that it didn't exist. That was the initial use case. And immediately, and this is like a lesson for all the entrepreneurs, is like if you solve a problem for yourself, well, there's like a million people like you. Well, turns out 50 million in our case [laughter] that would want to use your software. So that was the initial use case, and it was a really great stepping stone because we always got a lot of new users that are learning and are like wanting to build. And so think about it as like we're extending the platform as people are kind of walking on it, and you know, you might fall off because you're doing something, you know, that's scaling really fast, we can't keep up with you. I'm sure Supabase kind of felt the same way. But at some point the platform became complete, and then people could build real businesses on the platform. That was when we felt like, okay, now we're a real company, and that's when the revenue started coming too.

Yeah, there's two impacts to that. One, the number of people, I think like 80% of people who work at Replit have learned to code on Replit or use Replit. Two, like a lot of big enterprise deals that we do, the champion there is someone who's used Replit. And three, a lot of CEOs, their kids learn to code on Replit, and so I meet CEOs all the time. That's why I'm wearing a suit. That's the price of success, obviously. It's coming for Ant too. We still got some ways to go [laughter]. We're talking about when this conference will be suit and tie, but hopefully not anytime soon. But you know, meet a lot of big companies, CEOs. We're starting to do these big deals with companies, and yeah, they have memory of Replit because their kids learned to code on Replit.

Yes. Kids were coding on their phones in class. And we've had a couple of like high-profile kind of blowups where a kid in India like started missing school, and his mom went on Twitter and said like Replit is corrupting young children and he should be studying for his IIT exam or something like that [laughter]. And it like went super viral because it became like a discussion about parents in India and like the pressure they put on kids and whatever. That's one. Another was there's a kid that built a portal to go into the internet and get around firewalls at school. And there was like four or five different like really public cases where, you know, student hackers were like doing really important and cool stuff. Some of it is like, you know, maybe a little shady. And then it turns out there are IT providers that control the firewalls across the US, and they block wholesale, and like our usage went down like crazy. We ended up talking to them and like, you know, maturing the product a little bit more, implementing some firewalls and getting unblocked eventually. But yeah, it was a scary time.

I'll tell you why. One, like you should be, as a developer, you should be running 20, 30 agents at this point. And if you're running 20, 30 agents, even like a MacBook with 128 gigabytes of RAM, if you're running like a fairly sizable codebase, you can't run that many agents. And so a cloud development environment is going to be crucial, or you're going to, you know, be behind. The other thing is like supply chain attacks. We're seeing all these like high-profile attacks like Mker and Versell and others. And I think a lot of it is because people are using their local dev environments. The cloud will install a package on npm that's been infected, and then goes in your computer, and then use your credentials to republish on npm, and then like hack your company's internal network. Whereas in cloud environments, you can control them a lot more. For example, we have a package firewall. We use this company called Socket. I don't know if you guys work with them, but Socket is great. We do a bunch of other stuff like we don't let you install npm packages that are like a few days old because that's where the infected packages usually are. So I think for all these reasons, you know, I would have expected... but developers are some of the... it's interesting. We build all the world's technology, and we're actually like pretty late adopters if you think about it. Developers tend to be late adopters. Think about Vim, or like I'm sure someone here still uses Vim. Is there a hand? Here we go. There. [laughter] So...

I saw, but I used to. [laughter]

Not necessarily. I mean, we have this shared success story with Magic School, right? Magic School. You know, this teacher in Colorado quit his job during COVID and like was doing some soul searching and decided to build an AI company to help other teachers, you know, automate parts of their jobs. And he built it using Replit and Supabase. And it's like I think worth half a billion dollars now. It's like it's remote, I think based in Colorado or like hybrid. There's like a bunch of stories. We were talking about the UK earlier. We just opened an office there. There are a bunch of new companies like unicorns there, Granola and a few others. I mean, you're in Singapore. There's like a bunch of unicorns there. And then like remote actually works for a lot of companies. GitLab was an initial kind of really big success story. Supabase is the other big one now. And so I don't think it's necessary. There are a lot of good things about the Bay Area, but there are negatives. Like, you know, people who work in the Bay Area, they're always looking over the shoulder, like, 'Oh, maybe I should go get this other job. Maybe that's exciting.' And their friends are sort of egging them on or something hypy, and it's like they want to go follow that. And there are like events every day, and they're like distractions. But the cool thing about the Bay Area is that everyone's talking about the latest and greatest things, and so you're living in the future. You're riding in a Waymo. You know, there's an aspect of it that you feel like you're in the future.

Yeah, it was. You know, when we were like six, seven people, we were essentially living together in this like apartment building in San Francisco, and we would like cook together every night. And it helped a lot. You asked me about this 10-year journey. I think if you're not having a lot of commercial success, keeping people motivated is really hard. And the best way to keep people motivated early on is that they feel like they're working with their friends, and they feel like they don't want to leave their friends. And so, you know, that's one way to like get the culture really driving together and really good. You can't do it too much. As you grow, people start dating each other, and like all sorts of drama starts. But it's a great, great place to start.

Yes. I mean, I think so. We published this blog post called 'The Self-Driving Company,' and the idea behind this blog post is once you truly embrace agents, it changes fundamentally how you work. And so who contributes code? I think at most tech companies now, it's opened up to a lot more than just engineers. We actually have support people sending in patches, right? They talk to customers all day, know what to prioritize. Agents write most of the code. The engineers review them. And we built systems internally to safeguard that and put guardrails. And designers, some of our designers are the biggest contributors to the codebase right now. There's a lot of like CSS fixes you need to do every day and like tweaking to the UI and things like that that you don't really need an engineer to do. Product people at Replit oftentimes spin up new product experiences like Slides. Samuel, one of our product people, built Slides from scratch. And I was always skeptical of it. Like, I didn't start Replit to build Slides. Like, why would you want to go to an IDE and like make a slide? Turns out like now 35% of our artifacts people are creating are slides. I use it for slides. I was like, ah, it's a pretty cool experience for that. So they prototype new bets that like we wouldn't kind of prioritize for engineering, and so that's good.

Yeah. Yeah. So I think, you know, the way the technology gets adopted is typically like tinkerers, you know, broadly we call them 'proomers,' start adopting technology early on. You know, some of them start companies, and others like are just like tinkering as hobbyists. And then people start bringing it to their jobs, and they find use cases at their jobs that are quite impactful. So one of our earliest stories was someone like a tinkerer playing with Replit Agent early on and brought it to Zillow and drove millions of dollars of extra revenue for the company by optimizing something. And came to us and was like, 'Hey, this was a huge success. Want to roll it out company-wide?' And we're like, 'Okay, I guess we could do that. I guess we need salespeople.' [laughter] Luckily, we had SOC 2 because I remember thinking like enterprise is probably important. So I was like, theoretically, what do you need for enterprise adoption? I was like, SOC 2. My friend Christina at Vanta does SOC 2. I was like, 'Hey, I need SOC 2.' She was like, 'What do you need it for?' I think it's going to be important. So I've done a lot of like kind of silly decisions early on that made Replit good for enterprise. Our focus on security, like I have a cybersecurity background. I used to be a kid hacker. And at some point, but you know, our focus on sandboxing all of this stuff that we developed experience and talent in is extremely important in the enterprise because vibe coding is getting adopted in the enterprise, but in an unsafe way. So technology leaders are looking around, who's the most responsible player, and they're finding that Replit has the maturity. And so we've been like dragged into enterprise. But I will say there's something about it that has like a little bit more peace of mind. Like the contracts are a lot larger, there's expansion that happens quite good, the retention is a lot better. So you feel like the business is maturing.