He runs a $9 billion company whose engineers almost tripled their code output this year thanks to agents. So, can Replit run itself? That's this week on Platformer.
This episode is brought to you by Jira by Atlassian, where teams and coding agents get the context they need to do the right work. Try it free at jira.dev. That's jira.dev.
Welcome to Platformer. I'm Casey Newton. This week our guest is Amjad Masad, co-founder and CEO of Replit, which started as a website where beginners could learn to code in a browser and is now one of the biggest names in vibe coding. But it turns out he hates that term. We'll get into it, but first here to give us some data on the state of AI and jobs is platformer fellow and Gen Z AI correspondent Ella Marianos.
Ella, how are you today?
I'm doing very well. How are you?
I am good. Now, when it comes to vibe coding, have you ever used Replit or a similar tool or what is your vibe coding tool of choice these days?
I'm only doing stuff in Claude to be honest. It's the platform that's locked me in. And I'm never doing—I think the thing I use it for the most is creating apps that waste my time.
I was actually recently hanging out with a friend of mine who I discovered had independently developed the same addiction as me, which is building apps that let you create ELO rankings of things you like. ELO is the ranking system in chess, and so you just swipe between two things you prefer. I do it for dresses if I'm online shopping, so it gives me an extra recycling of the online shopping experience. My friend actually did this for all of his Spotify liked songs to get an additional metric on what songs he prefers the most.
And so you do this long enough and then at the end you essentially have a ranked playlist of the songs in the order that you actually prefer them.
Well, listen, I've heard of worse apps. I'll say that much. This seems like a pretty fun way to spend your time. And speaking of spending time, it's time to find out once again what's going on in AI and jobs. Not my best segue I've ever done, but it was what I had today.
I appreciate that, Casey. Thank you. So basically, the World Bank has a new report out about AI in developing countries, and it's kind of a change on the narrative that I've at least been seeing a lot on x.com, which I unfortunately still have a lot of contact with, where people are talking a lot about AI sovereignty, as in countries who aren't at the frontier developing their own AI models, and about how if they aren't at the frontier, they're going to be in the permanent underclass. They're going to get washed by OpenAI and become irrelevant. That's the narrative that the World Bank really disagrees with. In particular, the World Bank thinks that currently developing countries are better set up to benefit from AI development because fewer jobs in developing countries are susceptible to AI automation.
Okay. So, tell us a little bit about why they think that.
Yeah. So the headline estimate that they have here is that they've kind of put together an index of jobs that are very at risk from AI versus jobs that stand to see productivity boosted but won't necessarily be eliminated by AI. And for jobs that are at risk from AI in developed countries, they say 14.2% of jobs are at risk from AI. I might put that a little higher. But in developing countries, they say that only 4.5% of existing jobs are at risk from AI, basically because developed countries have more professional services jobs, more IT and communications jobs, and they think computer-based work is more vulnerable. But then the amount they expect to see benefits is about 16 to 18% both in developing and developed countries.
And what do you mean by benefits? Like is the idea that these jobs will become more valuable and so wages will go up or what?
Yeah, the idea is the jobs will become more valuable. So employment will stay the same and hopefully wages go up.
So yeah, so I mean look, we always love to feature optimism on the show and I'm somewhat heartened to hear that the World Bank has drawn these conclusions. At the same time, when I hear that only 14% of American jobs could be subject to automation, that seems sort of surprising to me given everything else that we've ever talked about on this show. So are there any nits that you would pick with this particular report or its methodology?
Yeah, to be clear, it's 14% in an index of developed countries, not just American.
Yeah. But anyway, one thing that I just want to impress in the mind of our listeners is when economists say estimates to you about what percentage of jobs are exposed to automation, by default do not trust them.
So the methodology behind this particular thing is somewhat complex. I want to be a little bit more careful with this one. First I'll say for background, the studies that are worse than this World Bank report that some other economists rely on, of which there are dozens, are studies where people literally—professional economists will literally ask the most recent ChatGPT, 'Do you think you can do this task? Is this task exposed to AI automation?' And then they will make their index based off of that. And please don't do that if you're listening. Please don't do that. Yeah. Thankfully, the people who the World Bank is basing this analysis off of have created a methodological innovation, which is that they have this complex pipeline where they survey people about a certain task, they survey experts about a certain task, and they survey LLMs about a certain task, and then they ensemble these things to try to get a reasonable, balanced picture. But I will say that while this is a 2026 analysis based on their idea of what tasks will be automated, the original study that tries to figure out what tasks are vulnerable to automation doesn't use LLMs after GPT-4o.
GPT-4o, I was using that in elementary school.
Yeah. Yeah, I mean all kidding aside, that is a very old—and I think if you haven't updated your sense of what jobs might be prone to automation since GPT-4o, I don't really think you're part of today's conversation.
Yeah, this is like really the problem. I can't even necessarily blame the people doing this analysis because it takes a really long time to make these studies. They had to wait on survey data from many thousands of people. They had to go through peer review. They had to ensemble these things, do a bunch of complicated statistical analysis. And so they just don't have time to make an analysis of the quality that they're making and have it be the most recent model. But this is the problem: even if you're listening to the best economists, this is what's going to happen to you, because it just takes them so much time to put together these types of studies.
Yeah. Something else that might make your statistical analysis take longer is if you're using GPT-4o. So I hope that they've at least upgraded to, you know, Sonnet 5.6 since this work began. Last question on this, Ella. Something I've always wondered. Does the World Bank have ATMs?
To be honest, I did not find that out while researching for this segment.
Well, listen, if any listeners know, we're curious because I need to get out a little money before I head to the airport. But before we do that, Ella, thank you very much. Very interesting study and data as always. A lot more to say in the conversation that is coming up about the potential for AI to automate jobs away. Somebody with a very interesting perspective on that. The CEO of Replit, Amjad Masad, coming up right after the break.
Jira by Atlassian is where agent speed meets team intelligence. Assign any work item to your favorite coding agent and see exactly what it's doing without leaving your flow. When something stalls, you're not digging through logs. You can see what's running, unblock what's stuck, and stay in control at scale. Don't let your agent start cold. The teamwork graph feeds them context from across your entire stack, delivering 44% more accurate results with 48% less token usage. So the work they pick up is the right work done right the first time. Try it free at jira.dev. That's jira.dev.
My guest today is Amjad Masad, co-founder and CEO of Replit. If 2026 is the year that vibe coding went mainstream, Replit is a big reason why. The company shipped one of the first commercially available AI coding agents in September 2024 and has ridden the surge of interest in coding tools to become a real leader in the space. Amjad grew up in Amman, Jordan in a family with Palestinian roots. He was writing software as a teenager and selling it to local internet cafes. He came to the US to be a founding engineer at Codecademy, a company that teaches people how to code, before going on to run JavaScript infrastructure at Facebook. In 2016, he founded Replit with his wife, who is the company's design lead and also works with his brother. The product started as a place to run code in a browser without having to install anything. Today, it uses the more familiar chatbot model. You type a sentence and an agent designs, writes, tests, deploys, and hosts an application. In March, the company raised $400 million, led by Georgian, at a $9 billion valuation, which is triple what it was worth six months earlier. Replit says more than 50 million people build on it, including 85% of the Fortune 500, and that it's on track for a billion dollars in ARR by the end of this year. That's up from $28 million in revenue in all of 2024. And in July, the company published a really striking claim, a blog post that said its engineers nearly tripled the amount of code each person ships just over the past six months and said the quality of the code stayed consistent and that the company can now close its customer support tickets 60% faster than it used to. All because agents are now working alongside engineers at every step. That's why they're calling it the self-driving company and they say that their people don't feel automated, they feel promoted. The company has also had to fight its share of battles against competitors like Linear and Cursor, and also against Apple, which for a time blocked Replit from being hosted on the App Store. And since you could argue that they compete with Anthropic as well, here I offer my usual disclosure that my fiancé works there. So there's lots to talk about. Here is my conversation with Amjad Masad.
Amjad Masad, welcome to Platformer.
Thank you. I'm a big fan of the show, so I'm happy to be here.
Thank you. Well, so I want to start with a vocabulary term because Replit, I think it's fair to say, is one of the biggest names in vibe coding, but I know that you don't like that term. So, tell us a little bit about why that term grates for you.
Well, look, I mean, it was never about coding. It's always about solving problems. It's about creating things, creation, making software, making apps, making exciting experiences, making games, making all of that. And the whole mission of Replit was always to make it so that anyone can get access to this technology. This technology has always been accessible to 0.5% of the human population, those who could afford a computer science degree. And when you put coding in the name, you're immediately signaling to people that they have to learn something complicated. Of course, you put vibe in front of it and maybe that makes it a little more digestible, but still, I think I talk to a lot of people and they're just like, 'Yeah, I'm not into coding.' 'Well, have you tried vibe coding?' 'Well, yeah, it is coding. I'm not a coder.' And the other thing is all of AI is vibe coding at this point. AI, any work app, Cowork, all these things, they're writing code under the hood. And so what we realized at Replit early on is that coding is the substrate. It's the infrastructure that makes AI really powerful. And so we'd rather think about Replit as a general problem solver, as a creativity engine.
That makes sense. I feel like if we're going to defeat vibe coding, we do need an alternative term. And I wonder if folks have suggested any to you or if you've tried out a few on your own.
I have not. I hear all sorts of suggestions from people. Nothing really stuck with me. But I think it's one of those things that will kind of disappear into the background and just become about creation. Like, I want to make something and you think of Replit, you think of apps like that.
It makes sense. I mean, it's sort of on my mind as a non-technical person who now has two apps on my Mac that I made myself by typing into a box. It feels really cool. I almost want to go in the opposite direction and just tell people I'm a 10x software engineer. Just albeit a non-technical one.
So interesting you're saying that. I think not only that, the thing is not only a 10x software engineer, I think in the future you're going to be a 10x mathematician. I think you're going to be a 10x biologist. I think you could be a 10x machine learning researcher. So take me for example. I've always been into AI theory and how AI works and the future of AI and all the sci-fi around AI, but I was never a researcher. I wasn't trained to be a researcher. I don't know how to construct a neural network from scratch. I hired great people who could do that, but I myself lacked that skill. And so I was like, okay, AI has gotten really good at software. Can it do other things? And I'm seeing people use it for math and all sorts of things like that. And so I was getting into chess around the same time, which is a very old ripe age, 38, of getting into chess. You know, you get into chess when you're six years old. I've been getting humiliated by three and four year olds online. And I was like, man, this is not going well. So let me actually do what I do best. And I created a chess engine. And the challenge I wanted to give myself was can you create a fully neural network-based chess engine because most chess engines are hybrid like AlphaGo, AlphaZero. It's sort of like part neural network, part traditional programming and search and things like that. And if you look at the literature out there, and I kind of used ChatGPT, whatever, look at all the literature. It's actually kind of an unsolved problem. I was like, okay, let me take on this challenge. And I was able to train a chess engine and I put it on Lichess, a platform kind of like chess.com. And it's now playing against humans, playing against other bots, and it's got an ELO score of 2300. And when it's playing against humans, actually, 1700. And so I'm like, wow. It's sort of like in The Matrix when Neo—they ask him, 'Do you know how to fly an airplane?' And he kind of lags a little bit. Like, I know how to fly a helicopter.
I feel like I have the same power. Like whatever you get—I have this genius intern, but that's also really sloppy and bad at ideas and things like that. And I have to direct it in a certain way, but I can do anything. It's really fun. And I feel like I've had similar moments myself as I've dabbled here. But let me go back in time a little bit. Because you've said publicly that you think it basically no longer makes sense for most people to learn how to code. And I think that's interesting because you were a founding engineer at Codecademy, a company that taught people exactly that, right? How to code. So tell me about your journey from teaching people how to code to deciding that they shouldn't bother. What was the moment that you said this old way of doing things is over?
Yeah, I mean my journey in thinking about coding education starts really young. I learned a bit of programming when I was seven or eight years old. My father got us a computer when I was—it was very unusual. I grew up in Jordan in the '90s. There weren't any computers around me. So it was like one of the first computers in my neighborhood. And so I built a business when I was 13 years old. I was selling software to internet cafes. You know, internet was expensive. You had to go to a place to use the internet and they didn't use software to have accounts or protect the machines. People could download malware and things like that. So I built this entire management software. And the thing that really annoyed me is dealing with the minutia of programming. Like, I made this great application, how do I get it into someone else's machine? So deploying it, running into basic errors. And I was like, programming is really fun and could be a great way to become an entrepreneur if you're someone who's interested in impact and making money and all of that. But also there's all this annoying aspect of it that should kind of be automated. And so that's kind of the spark of the idea. And then I ran into another problem in college. I didn't have a laptop, so every time I'd learn a little bit of programming, do some homework, I have to set up the programming environment over and over again. And so luckily you started programming after AI and AI could do most of the setup, but back then downloading Python and Java and all the stuff and you have a missing DLL or whatever, it was a huge headache. So I was like, okay, why can't I open a web page and start coding? And so that was the initial idea for what would become Replit. So I built the first open-source coding engine that could run in the browser and that went viral on the internet. And that was an insane moment for me because here I am, a kid in Jordan hacking in my parents' basement and I'm looking at Hacker News and it's the top story on Hacker News. People talking about it at conferences.
And then there's this little YC company. Literally before Demo Day, they went viral. They were on Hacker News, they were on Twitter, they were in TechCrunch, all these different places. And I look at it like, man, that kind of looks familiar. Well, it turns out they're using a lot of the software that I'd written. And so they gave me an O-1 visa. I become the founding engineer. I move to New York and we start working on this company. And the kind of stories we hear were amazing. It's the kind of stories that I hear every day now, but back then we would hear them once a quarter, which is: I learned a little bit of coding on Codecademy and now I used to be a fitness instructor and now I built my fitness app and now I'm making $10,000 a month and I feel like I'm financially independent and all of that. I was like, wow, this could be really impactful. But still the coding aspect of it fits a certain personality. You have to have a lot of time. You have to have a high tolerance for pain. And no matter how much we made Codecademy easier and coding easier, it was still not attracting the general population. And so when I went to start Replit, the idea was let's blur the distinction between learning and building. You should be learning as you're building because learning in isolation is kind of boring. I was never a textbook kind of guy. I actually did kind of poorly in school. I always learned by doing projects. And so Replit's core thesis was that there shouldn't be a learning step and then a building step. You should just go straight to building.
Now without AI that's actually incredibly hard. We did a lot of more classical automations around generating code for you using wizards and things like that. But I always thought that the key to solving this was NLP, natural language—no one says NLP anymore, but LLMs came out of NLP, natural language processing, field of research. And so when I saw GPT-2 in 2019, I was like, oh, we're getting close. We have these new machines that could understand human language. Surely we can translate that into code. And so we started building on top of that. And I wrote an article actually, it's on a blog in 2020, and I said, we're on the verge of solving programming. I think for the foreseeable future, for my vantage point at the time, you still need to learn to code, but you have to learn less and less of it. And I actually came up with—I was on the YC podcast, they called it Amjad's Law, but I said the returns on learning to program are doubling month over month. Which was true up to a point. Now in 2025, when we started getting Opus and the new Claude models, I was like, okay, we're getting to a point where I need to be honest with myself. My entire life mission around teaching people coding and making coding easier is changing. You don't need to code anymore. And I started building things without looking at the code and I was like, okay. This was before Karpathy even coined vibe coding. And so I was on TBPN and I said, you know, it was actually kind of a painful moment for me because here you are, you dedicated your life—I'm 37, I've been working on making coding easier for literally 15 years, most of my adult life. And I'm saying everything that I work towards is kind of moot. And the way forward is that we make programming so easy that anyone could do it and you don't need to learn to code anymore. And so I said that and that was hugely controversial. A lot of engineers on Twitter and Hacker News kind of attacked me, like how could you say this? It's not true. You're going to get people in trouble because they're going to build apps that are insecure. I was like, well, any problem AI could create, AI could solve. We could create a security agent. None of their concerns were actually unsolvable. And lo and behold, here we are mid-2026 and most engineers, not just vibe coders like you and I, most engineers don't look at the code. I talked to engineers at our company, they kind of only look at the code when it's about to get deployed. They do a code review and increasingly we're automating the code review process as well. And so it was this big painful moment, but I think as an entrepreneur you have to be honest with yourself, otherwise the company would die and you would not be aligned with the future.
Totally. I mean, it is amazing to me through that entire story how much has happened just in the last 12 months, right? Because I can remember when Claude Code first came out and thinking, should I get this running on my laptop? And disclosure, my fiancé works at Anthropic and he basically said it's not worth it, it's going to take too long, it's just kind of a whole pain, it's probably not ready for you yet. And then you fast forward to November and all of a sudden you basically just type 'let's go.' And all of a sudden, I can skip the last 15 years of your life. So I can understand why that would be super painful for you, but it's also great for me. And obviously, you guys are thinking really hard about how to turn that into business opportunities for yourself. And speaking of that, and at the risk of getting ahead of ourselves, I wonder how much of this conversation applies to design. In June, I interviewed Figma CEO Dylan Field about AI and design. According to Dylan, designers have never had a brighter future. You all just released Replit Design, which promises to automate a lot of the design process. So, is design about to hit that same moment that coding did? And what do you think it means for people who are designers today?
