We have CEOs that finally feel unleashed, like they have Replit and they have an idea. They don't have to go beg someone to do it. They can just vibe code and bring it into a meeting and say, 'Look what I built.' We want to get to a point where you don't have to code at all. You should be in a creative space. A lot of coding is minutia. A lot of coding is accidental complexity. And turns out there's actually research: doctors who play video games have much better reaction times. So if you want to do a surgery, always ask them if they're a gamer.
Oh, not a gamer. Sorry. But where's the gamer here? Get the gamer.
AI is reshaping how we work, learn, and create. And nowhere is that change more tangible than in coding. That's where today's guest comes in. If you've heard of Vibe Coding, you've heard of Amjad Masad, founder and CEO of Replit, the platform that lets anyone build software right from their browser.
Amjad grew up in Jordan, where he taught himself to code before studying computer science and moving to the United States. He became a founding engineer at Codecademy and later led JavaScript infrastructure at Facebook. What you might not know is that he's also an empathetic, people-focused voice in the AI space, talking about issues and philosophy that others won't. I've used Replit myself and it's clear to me this isn't what's next, it's what's happening right now. Replit makes building software as natural as writing an email, shifting the power of creation to anyone with an idea. So yes, this is an interview about coding and the fastest-growing programming language in the world, English. But even more broadly, it's about how AI is already forcing us to rethink how work and society function, told through the lens of one field that's transforming before our eyes.
And we're thrilled to welcome to the show, Amjad Masad.
Amjad, I'm so glad this worked out. We've known each other for a number of years. And part of the thing I love about doing a podcast is people that I learn from. And there's been a whole stack of things out of our conversations, you know, mostly at the Grove, but other places as well. So, welcome to Possible. It's so glad to have you here.
Thank you for having me. I'm really excited. I've been listening to the show recently. So, happy to be here.
I'd like to start with a question about kind of you, right? Like, you know, Replit's about making things. And so you might even phrase this question of how are you made, you know, tell me about your early experience playing video games and how video games kind of did your trajectory into where, and then we'll get into more depth in Replit, right?
Yeah. So video games have been a huge part of my life. I started playing video games I think on the Atari, maybe even before getting my first PC.
I wonder how many people still know what Atari is.
Yeah, maybe we have to actually describe what an Atari is. But here's a bit from history. Steve Jobs worked at Atari. It was like the Silicon Valley premier company for computers. And so I just was captivated by video games. And when I first got a computer, my first idea was like, can you actually make a video game on this thing? Well, turned out it was initially very, very hard. But I remember going every year back in Amman, Jordan, where I grew up. There's a computer show that my father would take us to. And it's like a really big place where they have all sorts of new computers, new devices, CD-ROMs, whatever. And I bought a CD-ROM in a collection of CDs. And I went home and I didn't know what's in them. One of them was like a fax program, whatever. I just bought CDs. I just wanted to try one. One of them was like a visual programming experience that allowed you to build kind of game-like experiences. So, one of the early games that I built was for my younger brother, five years younger than me, to teach him math. I think I was maybe eight and he was three or something like that. And there'd be two boxes. It would be an equation. He'd have to enter the right number, like, you know, five plus blank equals six. And it's like one. If he gets it right, he gets an applause. He gets it wrong, he gets a boo. And it worked out so well. He now works at Replit. So, I taught him math early on.
It's very early recruiting.
Early recruiting. Yes, early recruiting. Worked out. The moment you give a user a program you made, it's like horrific because they'll put in things that you don't expect. Like instead of one, it'll put an E. And I was like, 'Re, no edge cases, little bro.' Exactly. But then just the high of not only making something, but seeing other people getting value out of it just hooked me. And then video games just been through my journey since then. I got really into Counter-Strike and I used to go to these LAN gaming cafes and we would play Counter-Strike against each other. But all these cafes were kind of run by, you know, pen and paper. So you would go there, you'd pay for an hour or two. They'll write your name and what time you came in and they'll watch which computer you're in and like they'll tap you on the shoulder when your time's up. And I was like, that's so silly. Like there's computers all over the place. I'm gonna go write a program for it. And I was maybe 13 at the time. Took me two years, but I built this like client-server application to manage the entire thing with security and everything. And you could have user accounts and gift cards and I started selling it and I got really, for the time I felt I got really rich. I got all my classmates. I took them to McDonald's when it first opened in Jordan. And so yeah, gaming has been like a recurring theme to kind of pull me more and now Replit has a really fun gaming experience that you can build games with it.
So, you know, starting with your younger brother, how much is like a gaming background part of kind of the recruiting target, the cultural organization target, the how do you talk to people in Replit? In addition, of course, the dynamic of how the product operates itself, but how much does the gaming background inform that whole stack of the company?
Yeah, it was never explicit, but I think for whatever reason like the way we design the product, the kind of people we attract, the kind of art we put together around the product, even the product itself kind of attracts those gamer mindsets. And actually, you know, I've always intuited that people who played games early on had a, you know, their clock speed was higher. They could think faster. They could react faster. And turns out there's actually research around it. Like doctors who play video games have much better reaction times. So if you want to go to a surgeon, you want to do a surgery, I always ask them if they're a gamer. Oh, not a gamer. Sorry. Where's the gamer here? Get the gamer. And then when we were designing Replit Agent, which is our vibe coding tool that codes on your behalf, I was playing this video game called Hades and a lot of my colleagues got into it as well. And so in Hades, it is like a roguelike game where every game you start starts with a different run and it's actually a randomly generated game essentially. And so it was a very important mindset when designing AI experiences because AI is stochastic. And so even the language that we use inside the company, like an agent run is sort of like a game run and so on and so forth. So it's a big part of the culture.
So you've already described kind of your very first intense code experiences. It's the your younger brother. It's the manage the whole gaming LAN. How would you take the current kind of coding enablement, like what if your younger self had that coding enablement? How would your younger self have done this differently?
Yeah. You know the interesting thing about programming and you know this very well actually, I think my generation is the last generation where the tools were really a joy to use and then it became very industrialized. So when we were programming, like I started on my real programming experience when I built that business was in Visual Basic. In Visual Basic was a very creative interface. At some point I think in the mid-2000s like everything became, you know, this like Linux terminal, even building a JavaScript app. You have to download all sorts of software and hook them together. And so at least I kind of grew up in a generation where coding was like a little more creative. And then because it became such a high-paying job, it kind of I think changed because of that. And so as we're building the product, we got a lot of criticism early on even before AI that Replit abstracts too much away from the environment. And we're like we abstract it but it is there, like you can peer under the hood and I think that's what's important. But if you make programming so complex at the top people lose interest and that's why you know in CS classes in universities but in schools as well you can turn off a lot of people because they think it's this laborious IT task of like, you know, pip install and all of that stuff. Instead what you want to do is make things. So for kids nowadays growing up, for parents encouraging their kids to code, just go on Replit and make things and I promise you your kids will learn programming by osmosis. Like they can just see, they're going to get curious at some point. They run into a bug. They're going to go Google around and ask ChatGPT. And that's how I think the organic learning process happens.
And say a little bit about how the gaming dynamic enhances that. What's the way that the things you've learned from being a gamer to make that creative experience kind of come alive.
So there's no game in the world that starts with a manual. Right. So if your program is starting with a manual you've already sort of lost. So the first thing you want to get to is dopamine really quickly. So with Replit getting you the first preview, getting you the first thing that you can see that immediately just hooks you like, 'Oh, I immediately made something. I put in a prompt. I made something.' And then you want to create a safe environment where it is easy to go back, right? In video games, if you think about save and load, it's a very important part of exploring a game. You don't want to be in a position where you just die and that's it and you have to restart all over. So Replit we spend a lot of time working on checkpoint and restore. So every action you make in the IDE, and if I can get a little more technical, every action in the IDE is actually stored in a sort of ledger. So you make an action to the file system, to the database, to the runtime, all of that is stored in a ledger and then anytime you feel like you got into a weird spot you can always go back or the agent did something you don't like you can always go back to the previous one. So that's another principle and then there's the aspect of like sharing and having other people use your thing. So publishing should be really easy, inviting other people. So we designed the multiplayer experience to be sort of like gaming as well. And really I think the mindset goes a long way instead of thinking of the onboarding mindset is just thinking about it of how it's an unfolding experience of more complexity. The first experience. It's like sort of fractal.
Yeah. Mobile level one, level two, level three gaming dynamics.
What's some of the places where you've been most surprised by what this unleash of early creativity has enabled?
Some of the medical things were really interesting. Because it's so outside of the purview. I wouldn't have imagined it even. But you know there's a woman that had like a very rare eye disease and required to do certain exercises and she built an app that helps her do these exercises with her eyes every day. There's a woman in Korea and her kid has also like a very rare disease and she built an app to manage that disease on a day-to-day basis. So medical is one that's really fascinating. I think there's more to do there. And then something also so outside of my purview is like go-to-market and like sales people, you know, there's a role in these organizations that I recently learned about is RevOps, those people that are managing a lot of the data flow within go-to-market teams and building the tools and helping sales people do their jobs and those people tend to use a lot of SaaS software, a lot of other data sources but they don't have anything to connect all these things together to create applications and make their sales people more successful. So that's go-to-market. And then finally I think one thing that's been really, really surprising is CEOs and maybe that gets some engineers to hate us a little bit but CEOs that feel disempowered because they've delegated a lot of things. They don't have as much input on the process. So, you know, we have CEOs that finally feel unleashed like they have Replit and they have an idea. They don't have to go beg someone to do it. They can just vibe code and bring it into a meeting and say, 'Look what I built and why can't you do that in two weeks if I can do it in two days.' So, that's another interesting one.
I mean, you're talking so much about building and you also talking about sort of for kids bringing them the whimsy and the creativity and the fun. And as Reed knows, I hadn't programmed since freshman year CS, so a long time ago. My co-worker Parth sent me a Replit login a few months ago and he's like just play around. And literally within 20 minutes, I had built an app and like because I was so excited and because I could do it right away, like it's just so easy to get onboarded. There's no barriers whether you're a CEO or a 5-year-old or someone who's a little rusty, you can get right in there. And I think Replit is sort of so closely associated with what might be the Silicon Valley word of the year, which is vibe coding. And so you've said that you don't love that word. We all use that word. How would you define vibe coding? And is there are there other words that you would prefer to use when talking about sort of the full suite of what Replit does?
Yeah. Well, I've learned through my times in the Valley is not to fight the hype words. Sure. Like I remember when cloud was controversial. A lot of people hated cloud but it just stuck around right so although I have feelings about it we kind of embrace it in many ways. So what vibe coding is like it was coined by Andrej Karpathy and he was the head of AI at Tesla at some point and he talked about how when he's programming, he especially when he's prototyping, he got to a point where he's no longer looking at the code where he's typing in the prompts, a bunch of code comes out and he hits tab or accept and he just keeps going and try to make as much progress as possible before it stalls out. He calls it this is no longer coding because I'm going by the vibes like whether if the vibes feel right, if the app is running then it is working and so that's how vibe coding came around. Now I think that the reason that you know for us might not be really fitting is because you know a lot you can do coding but a lot of times you're not coding at all and I think vibe coding might scare some people away because the word coding is still in it. We want to get to a point where you don't have to code at all. You should be in a creative space, right? A lot of coding is minutia. A lot of coding is accidental complexity. You know, the fact that I know that null is an object in JavaScript doesn't add anything to my life, right? But I had to learn all that stuff. But I think there's going to be a generation of programmers that don't have to learn that stuff. I think that's fine. And that's the history of programming. Like when Grace Hopper, the inventor of the compiler in the 1950s invented the compiler, she was actually inventing a form of vibe coding. She said, 'I want millions of people to experience programming because it's such a wonderful thing and changes people's lives, changes businesses, and I want them to program in English.' And she said that way before Karpathy said that and then the machine code people were like pitchforks. 'Program in English. No, that's, you know, you can't do that. How can you understand, you know, how much data you're putting into each register, whatever, right?' But the history of humanity has been let's abstract away the details so we can focus on the more creative things. Now, there's always specialists. There's still people kind of looking at machine code and there's people doing C and there's people doing JavaScript and I think there are people that are going to be entirely programming in natural language.
That's a great segue to one of the questions I've been thinking about is like what is the new literacy is the highlight because obviously you know previously literacy was reading and writing and part of how we get to a massive elevation of humanity is through as broad-based literacy amongst the billions of people as we possibly can because what enables everything else, enables learning, enables cultural evolution, enables patterns of thought. Now with the no-code code, you know, kind of like the vibing as opposed to vibe coding or the, you know, whichever, that's a new form of literacy. So say a little bit about how you think about this new form of literacy and then also add in a little bit is like one of the things I think is I think actually in fact everyone's going to be having a coding co-pilot, whatever, code lab, coding, because just like that co is like that's now part of just the way that you would also write like an email or anything else as ways of doing it. So give a little bit of this kind of literacy and future of how everyone's going to be navigating thinking and information spaces.
Well, the first thing I will say is that soft skills I think are more important than ever. So for a while in the US at least we're all STEM, STEM is still really important but I think you know you have a lot of engineers going to work right now and they don't really are not very good at communication, at like talking about ideas. They're good at writing the code, translating some of those ideas into code. But I think now the way we see our product being used by product managers. For example, product managers are some of the best vibe coders because they're really good at taking problems and breaking them down into their constituent parts and communicating them very well to the machine and then the machine can build them. So whatever we do if it's like a question about education in our education system, it needs to get much, much better at like being able to take problems and breaking them down. There was a hype word like I think 10, 15 years ago called computational thinking and like people were building startups around teaching people computational thinking and I think that's still important because underlying all of this is still computational. So when you're describing an idea to an LLM to build it, you need to be able to talk about databases, you need to be able to talk about persistence, you need to be able to talk about lists and things like that. So if I'm designing a computer course for young people, I would focus more on the abstract concept of computation as opposed to again null is not an object type of thing.
Part of that to elaborate is not only is the baseline computation about what the three of us learned even if it goes all the way back to freshman year CS which it tends to be a very deterministic algorithmic data structures, memory management, other kinds of things. But also I think what's going to happen is we're going to bring in a set of other computational concepts. I think it'll be probabilistic concepts because actually most of like where LLMs and everything else is very interesting, it's probabilistic. I think it'll be maybe even quantum but a bunch of other things where compute is not just like 1 plus 1 equals 2. There's a richer sense of compute. Have you seen any of that yet with Replit or is that still kind of in the glimmerings of what's coming?
We see the problems of not understanding it. Yeah. We see the problems of users not understanding that it's a probabilistic system, it's a stochastic system, they expect results to be reproducible and all of that stuff. So 100% like I think the future is understanding these concepts really well. And by the way once you understand them, you can predict how these tools might work and you can go from one tool to another still have the same mental model understanding of it. In the same way that you know when I was growing up I really understood how computers work and that allowed me to build you know all these businesses and programs whatever, understanding how LLMs work is going to be very, very important. Now the problem is most people's experience is ChatGPT that is trained to sound like a friend, you're not getting raw access into the LLM in order to like figure out how to program it. And so we just introduced a set of AI integrations. So now if you ask for upload like make me an app that takes a prompt, generates an image model, audio model, puts them together or something like that. It's so simple. You don't have to do anything. Just put in that prompt. It will tell you we're going to use OpenAI for this. We're going to use Gemini for this. We'll take care of the billing. We'll build you on behalf of them. But I feel like there needs to also be a sort of UI. I don't think we've invented the UI for how to mix and match these LLMs and how to program them. So, as much as I am a big believer in natural language based programming, I think at some point as agents get better and they're like working most of the time in the background, we need to invent the new interfaces for how to program these machines.
Well, also I mean you said that most people when they think of AI, they think of ChatGPT. And so when I'm talking about my friends who are not technical or not at, you know, not in Silicon Valley, like that's their only thought about what AI is. And so they can't even open their minds to anything else because that's the one experience they have. And I think one of the things we're excited about is that with Replit and tools like it, we're going to open up the possibility to build to so many people. And so sort of a two-part question. One is what is the ways that we can have more people experience AI in this positive agentic like super agency way and then like what happens when we open up these opportunities for more people to build.
Yeah. So Alan Kay has this essay that's, Alan Kay, one of the computer pioneers, has this essay called 'The Computer Revolution Hasn't Happened Yet' and in it he makes the case that literacy changed the world. We had democratic revolution, scientific revolution. You couldn't imagine any of those without literacy. Now, computational literacy or people being able to use a computer to its best ability, which is being able to program a computer will change the world in fundamental ways. I think you need an entire generation of people to grow up using these tools. Maybe I'm showing my bias because I have a history and kind of working in education. I was founding engineer at Codecademy. We taught millions of people how to code. And I like to help governments. Now we did a deal with the government of Saudi Arabia where they want to put vibe coding as part of their curriculum like an entire generation of people actually experientially learning how you use these tools will change the world in ways we haven't predicted just yet.
One of my first public Replit experiences with Amjad was reproduce LinkedIn. Yes. LinkedIn. I saw that. Yes. And it was partially because it was kind of, you know, it was a bit of a mimetic hack and everything else. Part of what I think both the AI revolution does and also what Replit does is expand the zone of your imagination. Yes. About what's doable. Walk us through kind of like what's a really, I mean and obviously like it's not a, I'm not looking for a manual because just like your gaming point it's like a manual, but like what kind of imagination should people bring with them to start learning Replit and bringing an idea to reality.
Such a good point because we still suffer although you don't need an onboarding or manual we still suffer from the blank page problem. So we see a lot of, we talk to a lot of users that sign up to Replit and then sit there for a little bit and then leave and we're like what happened and it's like I just didn't know what to type in there. Of course in retrospect you know us in living in tech it's like are you prompt, what do you mean but of course you know it's a skill. Now before I go back to how we solve it I think after people go through this hump it's an amazing experience because you become possessed, you walk around your day and you're looking for software problems to solve. You're like, 'Wow, I have this new hammer and now everything kind of looks like a nail.' And once that clicks, it's pretty amazing. So now we're doing a lot of work at the top of the funnel, like figuring out what kind of educational experiences we give people. What do we tell them about it? But the first thing I would say is don't complicate it. Like, you know, have a very simple idea, an app that you would have liked existed, like maybe every one of us has an idea for an app that you can't find it in the app store right, like I want something to do this or that for my kids, having kids is a great way because I was just thinking that there's so many things you're like ah just organize my camp schedule, right exactly or I want this educational thing I want to be able to do this fun thing with my kids and then talk to it like you would talk to a person just say well I have an idea and like you just don't worry about the grammar, all that, you'll get better at prompting but initially just really get in and with any kind of idea if you want some sample ideas I would just say like basic things like just say I want to build a collaborative whiteboard and then just imagine if you've done improv, improv is like a good way to do it, just like word association, I want a collaborative whiteboard and then look around the room and you'll see a plant that allows me to create a garden or something like that. So really just be generative and you'll get an experience and you'll get a feeling of it.
Do you know sort of who is the typical Replit user and like what kinds of things are they building already? Is it oh 80% of people are building whatever SaaS software.
Yeah. Yeah. Yeah. I think most of it is AI software which is why we built this recent thing. It used to be that you have to go figure out how to get an API key from OpenAI or others. But people have AI ideas like they see AI and like a lot of people will be like, 'Oh, I wish ChatGPT did this one thing, right?' And then they want to build AI experiences like chatbots like ReadAI is a great example. Obviously, ReadAI is very advanced in that it has video and all of that. But you can build like, you know, I'm Jyoti, like go tell Replit to go look at your personal blog or paste in your notes and just say I want to create a chat experience that sounds like me based on these notes and so AI is a really big one. Another one is like people want to build businesses and this is you know as I started thinking about our business more there's two aspects of our business I would say maybe three, there is the consumer aspect which is you know building family stuff and building personal things and then there's the enterprise which is like automating things at work, line of business applications, you know, rapid prototyping, product development inside businesses, larger businesses and then entrepreneurs and the way I started thinking about entrepreneurs is that everyone has domain knowledge in their head that is not monetized just yet similar to how you know Airbnb found that everyone has space in their room that's not monetized, there are people all over the world that have niche specific problems that they've run into that no one else has solved. I was talking to an entrepreneur on Replit the other day. His wife is a yoga teacher in rural England and she does these popup yoga events and most of the communication is happening over WhatsApp. Payment is really hard to do. They use PayPal, all these different things. And so he wanted to build a platform for her but other people like her. So she doesn't have a yoga studio that you can go to every day, but she does these events at backyards and things like that. So he's built a platform and he has users and he has other instructors and so every niche in our life, there are people that are experts in that previously were not entrepreneurs because they didn't have hundreds of thousands of dollars to go pay developers to do it. Now they can do it.
I love that example because obviously there's entrepreneurs that you're serving that are building, you know, million-dollar, billion-dollar businesses that serve hundreds of thousands of customers, millions of customers, but I think the like solopreneur, the woman who has a yoga studio, like I think of my mom who's a piano teacher and she's always trying to figure out how do I get my schedule straight, how do I get payment, how do I figure this out? And so there's so many like she's an entrepreneur, the person who runs a yoga studio, the person who has like a bakery outside their house, like those are the people who would never be able to afford fancy coders, a fancy website, but Replit can empower them to do so much more. So I love that and there's something about it that makes me really excited about the world to come. As much as I love Silicon Valley, we tend to flatten the world. We just like build one size fit all. And that means like you know SaaS software tends to have all these buttons and configuration just to fit every use case. But instead we're going to explode that and decentralize it around the world. And every community might have a special business that does one specific niche thing. And that diversity I think is exciting. And as part of that diversity and being fun, we'd like to play for you a short clip of one of the ways people have used Replit to build something important to them.
Okay, so great.
What if you turn boba into a game? I prompted a basic app and sketched out some boba. I made three different drinks, taro, milk tea, and matcha. I added logic so the pearls fall from the sky. If you miss three bobas, you lose. But the game was way too easy, so let's make it difficult and add some bombs and a scoreboard. I noticed the game was a bit glitchy and I wanted to try it on mobile. So, I'm at my favorite cafe. I added more colors, different fonts, and harder levels.
It's the best thing. People can code on their phones right now, which is amazing. Exactly.
Done yet. What other features should I add?
That's fantastic. That's I mean, and we didn't just select it because she added it to her bio. I guess we first started working on the phone, people said we're just like crazy. Like, because back then there was no AI. And we had to build a lot of things to make coding on the phone like a little bit easier. But I always imagined because back in school I remember that I had a Symbian phone. I was like, I wish I could program it under my desk, you know, and so it's really cool to see people doing that now.
So one of the things that as you dig into AI it becomes you know one the earlier question which is how do you broaden your imagination but also now like what's a good way of optimizing like for example you know obviously you can do things that have huge token counts and that's kind of like doesn't build a good business you know etc. might be fun as a one-off thing, but that kind of thing. What is, as it were in the game dynamics, the next level of compute thinking, right? Because it's like, okay, how does this work within the operational envelope of like AI enabling it, being able to figure out what kind of interface elements, right? What's some of the things you've seen there in terms of, you know, kind of context optimization, managing the AI token outputs, you know, helping people with that kind of compute thinking?
Yeah, I mean there are two aspects of this. One is strategy. You want to be able to do things that the big labs are not going to do or eat your lunch and like actually be able to build a good business. And then the other thing what can LLM do right now and what scaffolding do they need around them? Right? The scaffolding part is really important because this is where you can build the value on top of the LLMs. So the way I think about our business right now, we were in the business of training models. Now we use all the different commercial models. And I see our business is to build the milieu, the environment around LLMs. If you think about LLMs as a person, you want to create the habitat for them, right? So we built the best habitat I think for programming, right? So when we first did this pivot from like the human being of the programmer to the AI being the programmer, I took the same teams at the company and I told them now you're no longer built for humans, you're building for the AI. So the editor team built editing tools for the AI. The cloud infrastructure teams that built deployment tools for humans, they now build it for the AI. So I would look at any different vertical and I would think this way, you know, if you want to build like a financial agent just think about what kind of environment you can build for that agent. Now I see a lot of people for example like trying to put LLMs into Excel sheets whatever which I think works okay but can you build an Excel interface for the AI right like in a way it's textual it can use it, you know, based on that you want to hit the right distribution of the LLMs, you can pick any vertical and if you want to build like a really great business just think about building your first user is the LLM, think about building that habitat for LLM to live in to create like amazing experiences.
Amazing things with product, the unlock of creativity, the unlocks of what does that mean for more personalization, things that work for individuals, businesses, you know, creative thinking is great. So, let's talk a little bit about also kind of some of the business attributes and like one of them is you know, you just mentioned that part of what you've done is opposed to home-growing your own models, you're now using a fleet of models, but that obviously makes you somewhat dependent on them, makes you more copyable, etc. You know, business attributes. So one of the things that the people who listen to our podcast, a lot of entrepreneurs, you're navigating a business challenge that they need to learn from. So how do you think about that? What are the things that you're doing to build unique and enduring business? What are the kinds of considerations that go into that?
First thing I would say is user obsession. There's a massive platform called LLMs, right? That platform is useful on its own. People can use it directly. But for you to build a very valuable business on top, what kind of value you're creating, one is if you're really obsessive over certain types of users, OpenAI can't build for every type of user in the world. So, pick a user that you know very well. Hopefully, it's you. You've had that experience. You're a salesperson. You were going to build the best LLM for sales people, right? That goes into UI, UX, all of that. Now, second, and this comment I'll go a little deeper on the comment about the environment around it. So you know I said that people using Replit should feel like they're in a safe environment. It should feel like anything is reversible. Now that turns out to be a huge technical challenge. Now you could easily kind of hack around it or say you know we'll just restart the app or do whatever. But what we did is we built a transactional file system that allows undo. Like we have a proprietary file system that we spent two years building that every action is like an immutable part of a ledger that anything can be rolled back and rolled forward that ends up being like a time travel system. And now we're finding more use cases for it. For example, one way to make models better, more performant is by doing sampling because models are stochastic. Every time you run the model, you might get better result, you might get worse result. And if there's a way to verify whether the result is good or bad, you can sample from the model. So what we do is because the system is immutable it has this characteristic of being very cheap to fork. So you can take a file system, you can fork it 100 times, run the same prompt with different parameters on these different forks and then pick the best one, do it repeatedly. Obviously that costs quite a bit of money, but for the advanced users, they're willing to pay it. They will get the extra value. So do hard technical things on top of the models in the environment where the models are sitting right, you want to build technical advantage it doesn't have to be in the model itself because that costs billions of dollars there are huge industries around it you don't want to be training models at least not yet when you're building initially building a business but you can do a lot of other hard technical challenges on top of it to make the models work a lot better. That's the second thing. Third is reaching scale. Once you reach scale you actually have negotiating advantages like you can negotiate really large discounts with providers similar to cloud and you can pit them against each other and so on and so forth. So and venture capital is an amazing system to give you that initial money that allows you to get to scale quickly and then the business you know and then you can enter more of an optimization phase of the business.
So you have this great metaphor which is cathedrals built from bazaars and so you want to describe the tension between open source and closed product design and the bazaar is this messy collaborative maybe fun bright and that's where open source lives and then you have the cathedral and that is the top down that's sort of Apple where they live and you said that you want to build cathedrals from bazaars. Tell us more about what that means to you.
Yeah so I think that comes from Eric Raymond or some of the original Linux hackers, cathedrals versus bazaars and there was a huge discussion early on in the world of software and so I gave a twist on it is cathedrals made of bazaars. Now in the early 2010s when this idea of no-code kind of emerged there are a lot of different approaches, you know, there are amazing businesses built there obviously, you know, Zapier, Airtable, Notion, you can all think of them as no-code builders. Now our approach which for a while didn't work until AI kind of really made it work. The amazing hyper-competitive diverse landscape of open source of people building all these open source packages and components. What's happening there? It's the bazaar. It's messy, but it's very innovative, very can you build a facade on top of all of this that makes it feel as good as a user experience as say a Notion? Wasn't obvious that you can do that, but the kind of abstractions you have to build on top of that and you can still hook into the latest and greatest.
Like when a new package comes on open source, Replit has it on day one because we built all this abstraction to be able to go get the packages and install them and run them on the system. And so we're riding a huge wave. We're riding multiple waves. LLMs getting better. We're also riding open source getting better. New programming language comes out, we can run it on day one. And that required a lot of investments up front. And there are a lot of competitors that are not doing this. And it might be short-term easier for them to do, but I know how the landscape of programming works. Things will shift. Like the other day, we saw a new JSON format called Tune that is more efficient. It's I think a token-optimized format that is more efficient for LLMs. And so again, programming will continue to improve and we want to be the platform where you can program anything with LLMs.
Let's think a little bit also about kind of how the future of work is going to evolve. You know, most people's experience right now is with chatbots. So like they put in something, they get out something, they put in something, get out something. So they're kind of directing the work in some ways but it's kind of iteratively from what they're doing. So how do you imagine this trend, Replit, others, in this kind of everyone working with, you know, kind of that power of code because you need a combination of that individual creativity, imagination, you know, broad imagination, drive, but also standardization for work processes. It's one of the reasons why the classic industrialization is I fill out this form and then I hand it to you and you know, D is way doing it. So I'm certain that you've been thinking about what this means for the future of corporate work. What are some of the observations you might share?
Yeah, when I first came to the corporate world, one thing that I found extremely depressing is how atomized and siloed people are. So, you know, over time, with capitalism sort of emerging, you have the original, I think Adam Smith, like how do you make a pencil, right? Like someone makes an eraser. And so, people become sort of cogs in the machine. And I don't think that is conducive to the human spirit. Like I think humans are fundamentally creative. I think anytime we assume the shape of a machine, I think people are actually not happy in their lives. I'm not a Marxist, but I've read some Marx and Marx's theory of alienation and the theory that I think Marx was right on a lot of the critique of capitalism. Maybe not the solutions.
