33 - Network State Conference 2025 - Fireside with Amjad Masad of Replit and Balaji

All right, a quick vibe check.

Has anybody in here ever vibe coded some hands?

Can you make some noise if you gave it a try? All right, well, with that, it's my.

Pleasure to welcome to the stage the co founder and CEO of the original.

Pioneering vibe coding firm, replit, Amjad Massad and Balaji.

All right, I'm Jad. I think. Is this your first time in Singapore?

First time in Singapore, yes.

Beautiful country, I think.

Love it.

Yeah. Awesome. So, you know, I think it was one of the. One of the early seed investors in Replay, I think. Right, yes, Many, many years ago. Yeah. See, and before that, you did react at Facebook and really one of the things that we've shared an interest in for more than 10 years is the developers of the developing world. Right. And helping people level up, you know, who are just getting access to a smartphone and so on and so forth. You know, my father came from, you know, very poor environment in India. You, you know, you came up yourself and I think Repl is really doing that. I mean, you've got some amazing stats now. What are your numbers now?

Like, we have about 40 million users around the world. We have a lot of developers that are starting to monetize their apps, reaching millions of dollars in the matter of weeks.

We have this crazy graph that you posted.

Yeah, yeah. I mean, it keeps going crazier.

Yes. It's obsolete already.

Yeah. You know, in about 10 months we went revenue from like 2 million to 150 million ARR. And I think it's more impressive when I think about this, it's like, oh, we're bragging about how Replit is growing, but what's more impressive is the amount of value that people are getting from the platform. And so when I think about when I see, for example, there's a VC in Europe that he's more like a CFO of VC firm and he's always been struggling with how to manage his portfolio and the CRM and all the data there. He decided to go on Replit and build something to solve his problem and then realized that it's actually generalizable, went out and tried to sell it. Before he quit his job, he had 5 million euros in ARR. And so just the ability of people to be able to create businesses on the fly. And, you know, this is something you and I connected on long time ago is the, you know, how the world is just going to be changing and startups are just going to be a lot easier and that, that changes everything.

I mean, I love Repl. I mean, Replit has created definitely billions of dollars in surplus value for the world. In that sense, you're, you know, the other definition of billionaire. You create billions in value for people. So it's amazing, amazing thing. I love it. For data analysis and for little, you know, these JavaScript Y math things that are really, they're, well, they're relatively easy to describe. For example, to just approximate PI by taking a circle and then having an inscribed triangle square, you know, then pentagon, hexagon. And as N increases, that actually becomes a closer and closer approximation. And I can just say that and it'll cook it up and it'll be a demo with a slider and so on. It would take me like a whole day with coffee to code that before. Now it's just done like that. And I love it so much because I can do that and you can cook up these amazing demos. And so it's just like a delightful thing that makes your life better.

Yeah, I mean, we spent 10 years, I mean, when you invested, we had started talking about AI, but we weren't sure it was going to be that nearby. But what we did. There's no platform in the world still, till today where you can go with nothing but an idea and go all the way to a deployed piece of software that you can scale. And so we're like, okay, what do you need? What kind of components do you need? You need the development environment, the package managers, the file systems, the IDEs. You build that, then you need the deployment environment, you need the databases, you need. And you build that, then you need the community and the social angle and the multiplayer. So we spent all this time building that.

Yeah.

And then a lot of the VCs, it's like, hey, why aren't you making money? You have a lot of users. It was like we weren't creating that much surplus at the time because, yeah, you're solving these problems and. But still, people get stuck. The hardest thing is just like coding the thing and managing the environment. And so at some point last year, we said, okay, our developers are no longer the primary developer. The primary developer is the agent that we're going to build. And so we took all these components that we built over 10 years and we created interfaces for AI. And now when you interface with Replit, you're interfacing primarily with agents. And like Balaji said, like, you don't have to have a grand idea. You can just like go there and, you know, for example, wanting to do some data analysis thing, you could just describe it and watch the agents Kind of work through it, and you can work with it collaboratively. So it's really come together in an awesome way.

There's like 10 different. I mean, one of the things that's interesting about repl.it is, you know, that seeing the overnight success 10 years in the making almost exactly because you had to do, you know, all the stuff you did with Nix and all the hard work with package management and so on. And I remember all of that kind of stuff. And that built this, like, strong frame.

And Balaji was a firm believer the whole time. I called him, like, last year, we had to do a layoff. We're not doing very well. I told him, hey, there's some people interested in buying the company. He's like, do not sell well. We're not doing so hot. And I remember you told me this. It's like, look, sometimes the outside view is more important than the inside view. Your view right now, you know, you're laying off people and, like, maybe revenue is not going so well, but I can look at it from the outside and see the entire ecosystem and see how important it's going to be. And so I'm always, you know, very grateful for that.

But you crushed it. I mean, as a little assist, tiny pass. And I think teamwork, you know, one of the things I want to do at some point, I taught this MOOC, if you remember, in 2013, the startup engineering MOOC, which at the time, 12 years ago had 250,000 students. I. You know, should we do one on replit maybe? Yeah, maybe we should. Yeah, that'd be.

No.

Yeah, yeah. Okay, fine. Great.

That's it.

Yeah, so that'd be pretty good, right? Because you've got millions of devs there, and I think we could have all the coursework optimized for it and we could have people level up all the way. And I think it'd be really, really powerful thing. We could also AI translate into all these languages. It'd just be like the ultimate kind of interactive environment. And then they would just come off of that. They'd be 2.0, you know, and I think, like, the thing about education of that kind is it's just incredibly positive sum, because actually, do you remember, you know, my favorite replica story? Actually, you may not even remember this. After ChatGPT came out in late 2022, we put up a replit bounty for the RoboJournal to automate the New York Times. And then there's. Do you remember this one?

Yeah.

And there was a guy who had just done like 90 days of code or 60 days of code or something. Right?

Part of our 100 Day of Code.

100 Day of Code, yeah, yeah. And so then he just crushed it in like.

Well, you should describe the backstory. The New York Times.

Oh, yeah, yeah. So automate the New York Times. No journo, only crypto. I mean, these are all basic things. The question was, for many years I thought that sports articles were wrappers around box scores, and financial articles were wrappers around tickers, and lots of political articles were wrappers around tweets in the sense of literally a function of those underlying things would produce article because it was very predictable. And I could see there was a company actually before chatgpt called Narrative Science. You never heard of them. So they actually were kind of a proto version of this where they took a financial report and they turned it into like color commentary, like a, like almost like a sports game. Like, you know, Nick's down three points, but this is like your Northeastern sales division is down 3% this month. So I could see that that kind of thing was coming. And then once ChatGPT came, it was possible. And then probably, you know, now I'm realizing that was like v00 of REPL. Vibe coding and this guy just knocked it out of the park.

Yeah. And I think the message there, and I think what's really important in the moment we're in and how it relates to network state and crypto is that when people have the ability to program computers, their reach on the Internet just increases massively. Right. It doesn't have to be. You don't have to just build startups. It can be activism, it can be freedom of speech, type. Type things that you can do. Or in that case, he sort of like making fun of how reporters write stories and showing the absurdity of the tech journalism industry. And I think that's really the power of. The crazy thing is we've had computers for 50, 60 years now, but we really haven't used them to the best of their abilities. Computers, I think, are the most powerful tools in the world since reading and writing. And it's going to create changes across society and civilization and education on politics and education. But it hasn't happened yet because programming computers is hard and full of like random IT side quests that you have to do. But if your Python environment breaks, the best thing you could do is throw your laptop in the garbage.

This is the one thing we may, I'm sure, if you still hold the strong view. Right. But I still think that it's worth knowing how this thing works under the hood and being able to build it on your machine. And I think a few years ago you said, do you actually have to know that? Do you still think that you know what I'm talking about?

Yeah, I mean, I'll give you redactio at absurdum kind of thing here, but like assembly code. Yeah, but see, when Grace Hopper, who is a computer scientist in the United States army, invented the compiler, she said, she's the original vibe coder. She said people would never want to program in machine code. We want to make it so that you can program in English. So invented the C programming language and the compiler and all these languages came after that. And everyone was up in arms against her, like, no, you need to understand machine code and whatnot. And then it's like, well, yeah, maybe some specialists need to understand machine code, but others just need to program. And. And obviously she was right. And then, you know, Guido and the inventor of Ruby came along and Brendan Eich and said, well, maybe you don't need to understand how to manage memory and do. Maybe we can, you know, build a garbage collector and you don't have to manage types. It's like, no, you know, you have to understand. So every move, every step was fought, every step in abstraction, in computing. There's pitchforks involved.

Yeah.

And I'm happy to receive the pitchforks right now. I'm not saying you bring up pitchforks.

So there's a thesis, I think there's a synthesis of our views which is as follows. So my counter argument to that is deep seek, which actually did need to go all the way down to the very low level bit twiddling type stuff to make its advances in it. So my view is everything you're doing is amazing because it increases the top of the funnel and it makes it. And many people just stop there, which is fine. Right? And it's great.

The cool thing about Replit, you can peel the layers.

Exactly. That's right. So you can keep going one more step.

I think that's the important part.

Like you don't have to.

Replit always generates open source code, always uses open standards, sh, postgres all that. You don't like Replit, download your code, go somewhere else and host it. And I think that's the important sort of principle layer is that as long as we're not obscuring how the thing works and you can peel the layer, you know, I think it's still healthy for people to be able to make something without having to learn how to code.

That's right. I absolutely agree with that. And the fact that you can peel it and lets you go. So one other thing actually related to this education. Right. You know, AI is both destroying traditional education in many ways because you have like teachers creating, using AI to create problems that students use AI to complete, the teachers use AI to grade and it's all just AI talking to AI kind of thing. Right. And, and so the entire kind of 20th century model is kind of melting right now. At the same time. A completely new style where the kids love replit, they can just type to it, talk to it, iterate on it. So what are your thoughts on that? Like, you know, what is AI? Take it the way and give it.

Yeah, yeah. How many people here know Bloom's two sigma problem? Yeah, maybe two, three people.

Probably the tutoring concept.

Yeah, yeah. People who work in education. Yeah. But basically there's a researcher that wanted to find out what kind of educational interventions we can make because the government every day has an idea for how to increase education attainment and things like that, but they fail, they always fail and sometimes they actually make education worse for the top performers. And so he did a research and studied every available method of pedagogy and things like that and, and found that the only thing that actually has an impact is one on one tutoring. And not only a marginal impact, a two standard deviation impact from the norm, from the standard lecture style education. And so that's been replicated. That's the most replicated study in a lot of the soft sciences. And so okay, what does that tell you? You know, you can be nihilistic about it. Right. It's just like nothing works. Let's go home.

Personal brainer.

What's that?

Like a personal trainer? You need a personal brainer for your brain.

Yeah. But the other thing that tells you is that AI is the ultimate solve for two reasons. One, the obvious one is you can do one on one tutoring for everyone. ChatGPT can be a tutor for everyone. Two, you can train new educators with AI. So there's a startup that started on Replit called Magic School and doing ed tech in the US is kind of the worst business ever. You know, I've tried it and usually it takes you 10, 20 years to get to 10 million. ARR, this company got to 10 million.

What were you saying? What was the worst business? Doing what?

So what they do is they help teachers increase their kind of impact and reach and velocity and being able to generate assignments and homework.

But teachers Are very tough market because they don't have any discretionary budget.

Right, exactly. But he created so much value for these teachers because us teachers are so overworked and under supported. And so now you can go to the magic school and prepare your assignments and do all of that stuff so you can massively lever teachers. That's one approach. Another approach, government of Jordan wanted to create like an AI tutor for kids. And they went to scale AI, they gave them $10 million price point for it. Someone there, an entrepreneur, decided to build it on replit, doesn't know how to code four days, got all the material from the country's curriculum, put it in AI, did rag and now students. They did it with a pilot of 1500 students and it was really impactful. And now they're scaling it to 2 million people. This is like one person built an app that is changing how the school system is working in a country.

One of the things also about repl.it which I love so much is it really does allow anybody to be, if they want to, they can be a CEO. Because actually like using replit agents is a lot like working with an engineering team or being an engineering manager.

Right. It's as frustrating as.

Well, I didn't say that far, but yes, sometimes, sure, sure. No, no. But my engineers, I love our engineers. But the part about expressing yourself very clearly about what you want to actually see built and your vocabulary to be like, okay, this div should be here and so on and so forth. And to be very precise about that and keep editing it and so on and so forth and the patient to kind of do that. It's really interesting to see that. Because most people cannot afford to hire an engineer.

Right.

But anybody can afford to hire a replit.

Right? That's right.

And so you've hyper deflated managerial training as well. Managerial education.

Yeah, I haven't thought about it that way. Yeah. And I think it has bearing on how we teach people. Like, I think the education system was built for the industrial world. It was built for like the drone worker. Everything about education is actually mirroring the factory model, the way we go through grades and all of that. And now everyone is a manager.

Right.

Like, so how do you change the education system so that you put a lot more emphasis on how to communicate very clearly, how to manage, how to be patient, how to break down ideas, complex ideas into different components, how to keep someone or an agent or an AI on track. It's a very different model than how we used to teach. And I don't think that many people are working on it. So if you're interested in education, you know, that's one area of research, potentially a startup you can work on, which is how do you actually teach people how to manage AI agents?

Yeah. So I have, I have a thesis on this. I'd like your thought which is the market is greater. Meaning like, you know, for example, at Harvard, great. Inflation has gone like just through the roof like this and everybody just gets an A. Why? Because they're basically like expensive customers of Harvard and so on. But ultimately they don't all, they're not all guaranteed to get a job. The market is the greater. So what if you pull that concept much, much earlier and you have kids start doing bounties and stuff much earlier and that's kind of back to the future. There's this great book called the Craft Apprentice, which is about the apprentice model in the 1700s, 1800s. That's what we had before the Bismarck Prussian 20th century school system. Right? Yeah. So now, I mean, the reason that that system got deprecated is because kids were working in factories and it was unhealthy and so on and so forth. So that's why the child labor laws and stuff happened. But just hitting keys on a keyboard, that's completely safe. They can be doing it parental supervised. So you kind of come back to the future of like apprenticeship on the computer.

And I think it teaches kids the true game of life. Right. Like I used to be gamer, love video games and I'm like very pro kids kind of gaming. But what is the ultimate game is creating and earning.

Yes.

And I think that teaches kids a, a lesson about, you know, how money works, how to save, how to, you know, run a business, but also just.

A life lesson and how the ikigai thing was that ikigai, you know, the four. It's like, do you like it? Are you good at it? Is it good for the world? Can you make money doing it? It has to be like all four because a lot of things that you know, you can like, but it's not good for the world or doesn't, you know, that kind of stuff. And this teaches them very early on to provide value. Right. So go ahead.

And also it teaches them freedom, responsibility. Like I remember when I, when I bought my first, you know, Xbox because I had like did some programming work and it's incredibly empowering and also just creates more free people because you realize that, you know, that you have economic power in your hands, whereas most kids kind of go all the way to being 25 years old before. Before they actually experience making any sort of money and makes them a lot more conformist because you know that your living is dependent on your parents than your school teachers, than your boss. And instead by just tasting that freedom of free market commerce, I think just changes how people think.

That's right. And I think repl.it could be. It's almost like a universal seed funding check. Because if you've reduced the cost of quote, hiring an engineer by 1000x or some number, right. Then that went from $100,000 to $100 to try whatever cool app or product you wanted to build. And you can almost quantify that with your billing numbers. And the models are going to keep getting better and the user is going to keep going up. So relate to that slightly switching gear. So you've got this huge user base. Huge repl user base. And we start to experiment with materializing replit into the physical world. And so we started to do some things you've done, obviously hackathons, meetups and so on. So we started to do this at network school. Right. Matt came over. I think that went pretty well.

Yeah, yeah, yeah. I think that, you know, there's so many things that the Internet has empowered individuals in, but you still need a community. And we ran the experiment with COVID where people went online and tried to start businesses and communities and all of that. And I think in many ways it taught us a lot. I mean there's really big businesses. Coinbase is I think probably the biggest business that is fully remote. And just like the idea itself, if you told someone in the 90s about it, it's just insane. Except the book Sovereign Individual talked about telecommuting in the 90s. But that was a really great experiment, but also taught us how important community and physical space. And I would also say that one thing I would really like to happen you and I discussed is like being able to program the physical world.

Yes.

Because I think we're headed to a period where robotics are becoming real. You can like buy a unitary dog or unitary robot. Like you should be able to vibe code it. Otherwise we're going back to the regime of being of buying devices but being dependent on third party to make the software for it.

So on that. Did anybody here play with the chess robots out there? Yeah, it's pretty cool, right? So we actually have a bunch of chess robots out there. Yeah. So I think the embodiment, you know, like, like a replit robotics, you know, the embodiment of AI is going to be very interesting because then the sort of managerial concept becomes real because you can like type instructions. They can kind of do things.

And I've been seeing this on Twitter and you and I, I think aligned on this is that, you know, there's this idea right now that we, we want to bring re industrialization to the US which is noble thought. But you're trying to build the world of yesterday. Why not build the world of tomorrow? What is the world of tomorrow is training people in robotics. We're going to have a lot of people that are going to be programming robotics to be able to create factories. The physical world is starting to become more programmable with work that you're doing with the network state. You know, being able to spin off communities.

Very Internet first. Yep, exactly.

So programmable spaces, programmable robots. And I think, I wouldn't say we're going to be focused on it in the next year or so, but, you know, definitely by 2027, I would really want to, you know, have REPLIT be your portal to your physical space as well.

So, you know, this point about like programmability or editability as I call it, you know, this was something that struck me where, you know, in, in your, you know, I've mentioned this before, but like you can build a billion dollar business from your computer, but in some places you need a billion permits to build a shed in your backyard. Right. And so the physical world is like read only.

Is what?

Read only, Read only. Yeah, you know what I mean? Like you can't, you can't move a stop sign, for example. Right. Lots of physical world is like fixed and making it programmable in some way. Well, there has to be like a test zone because you can't have too many people there. But if you can make the drones whiz around and the robots whiz around and you can tell the robots pick something up put together there. Like to your point, there's really cool videos and things now starting to emerge with like drone and robot construction.

Yeah.

Where they're actually like assembling something for you. You know, and you know, I'm thinking is, you know, obviously there's smart homes and stuff like that. What was your, what were you thinking when you talked about making the space programmable? What was in your mind?

I think we would start with fun things. We'd start with robots for kids. MIT Media Lab used to be this very great institution at mit. Seymour Papert, a huge innovator in education technology and things like that. He created the Turtle. I'm sure, you've programmed, some of you programmed with logo where you can move a turtle on the screen. The original turtle was actually a physical turtle. And he had this idea to truly teach people programming, you need a physical embodiment of programming. And he found that when children, what they do is they start thinking like a turtle, like the turtle. It's like I'm standing here, I want to do a circle. So therefore I need to move steps like that, I need to loop like that. And then they're able to learn programming better because of that. And so I think like, just like fun things, home automation things is where I'd start. But you know, I think, you know, ultimately if like I think the next generation of startups, AI is this bootstrapping of it, but the next generation of startups after that is, you know, making things that were hard to start. Startups around being able to start do it like, you know, SpaceX creating platform for space.

I mean so many space companies now. It's amazing. Elon created that whole category.

I think we need a platform for, for factories, platform for, for science. Like you should be able to spin up, you know, science experiments from, from your replit agents and, and so I think really the sky is the limit there, but we'll, we'll start probably small.

That's cool. Awesome. So any questions for me?

Questions for you? How do you think programming will play into the network state? I heard one of your employees say they think about the network state as a Docker image for new computers.

Yeah, it's a great question. We think of what we're doing with network school as a template for other startup societies. And ideally, ultimately I'd love to hit enter and take the bill of materials and have robots lay out the modulars, set up the solar panels, charge it from there, go from there.

And so how possible do you think that is?

It's getting much, much more achievable. Much more achievable. So like solar in particular has crossed a certain cost threshold where it's like really cheap. And I was never a big solar guy because I thought I was a big nuclear guy, but I've come around to thinking of them both. And you know, actually a guy who convinced me on this is a guy named Matt Green who's one of the co founders of zcash actually, which we'll hear from later today. And his point was, he granted all my points about nuclear and base load and solar would be wavy like this, they said, but solar has one huge advantage, which is you can just Rack and stack it. It's just really simple. And nuclear is complicated.

Right.

And it requires permits and so on. Right. And just that alone is like.

Yeah, worse is better concept.

Yeah, exactly. That's right. So there's a truth to that. Right. It doesn't mean you don't do nuclear. You still do nuclear. Because nuclear does scale. But it's a little bit like sandals versus shoes. Right, Right. This has a higher fixed cost and then a higher payoff. This one you just get started with right away. It's interesting, right? Yeah. So. So, like, the concept of a house that could run out of battery but is completely off grid.

Yeah.

Means you could rack and stack it horizontally in sunny environments. And there's so much that comes out of that that means you can just kind of start gridless. And people have talked about this forever, but I feel it's technologically suddenly become feasible. Like Pakistan. African countries have just racked and stacked Chinese solar over the last year.

Right.

This is an area I'm tracking a lot. We talk about programming the physical world.

Yeah. It's interesting how we're starting to use some of the same concepts from software in the physical world. Like, I don't know if you can talk about the modular stuff that you guys are doing.

Yeah. So we are doing what we want to do. Every. Here's. I'll give a quick summary and then we'll talk more later. So every cloud entity, imagine it takes physical form in a different way. Okay. So like, maybe Coinbase becomes a bank and replit becomes a hacker house. And so, like, each thing takes a physical form in a different way. And if we could put that spirit into a modular or a set of modulars, you could start replicating REPLIT and coining. Coinbase is all over the place. Right. So stay tuned. Maybe we'll do something like that. All right.

Yeah. I think just to, like, put a final note in this. Yeah. Like, I really want to be able to, like, drop a vibe coding lounge anywhere in the world. Like, you know, almost like from a press of button. I was like, oh, there's an event happening there. I want to have presence, but I don't want to coordinate with organizers, whatever.

For modulars.

Yeah. Yeah. Can you imagine, like, Davos, like, just saying, like, don't go to these boring politician talks and just like, go vibe, code something and like, drop a module there and people.

Yeah, it's very doable. And maybe we'll do it.

Yeah. Awesome. All right, thank you.

Thanks. Thanks, everyone. Thank you.

Pleasure.

All right, next item.