--:-- Paul Graham: I think with AI, everyone will have to go through this period of denial and then depression and then eventually acceptance.
--:-- Amjad Masad: Replit's like this iceberg where there's this visible bit above the water that's like, look, you just tell it in English what to do. And it writes your code. And then underneath there's all this stuff for deployment and sandbox.
--:-- Paul Graham: I don't think we describe ourselves as vibe coding now.
--:-- Amjad Masad: We think Replit is the platform for the transformation for it to self-driving companies.
--:-- Paul Graham: More than other companies though, I think you guys have been living in the future. If you had to say, what's the real DNA of Replit?
--:-- Amjad Masad: It is living in the future, for sure.
--:-- Paul Graham: So you guys basically invented Vibe Coding, right? And when you did it, like in 2024, which sounds like only two years ago, but it seems so long ago technologically. How did you even know this was going to work? Because it's obvious now that it did work, but at the time, it seemed crazy. It seemed crazy to me.
--:-- Amjad Masad: Well, for us, it was incremental. It wasn't like for the rest of the world, I felt like a big departure from how people did things. But we've always...
--:-- Paul Graham: You were already writing more and more of the code.
--:-- Amjad Masad: Well, no, it was sort of like the YC kind of ethos of looking at what your users are doing. And that will give you ideas on what to build. Because you were already doing autocomplete.
--:-- Paul Graham: We're doing autocomplete. So how much of the code were you writing before you went to full vibe coding where you tell the thing in English what to do? Where you're just generating code with autocomplete, how much of the code were you already generating then?
--:-- Amjad Masad: I mean, you could probably say 50, 60%.
--:-- Paul Graham: Really? So that's interesting. Even before Vibe coding, you were already 60% Vibe coding. And that was how you knew it would work.
--:-- Amjad Masad: But the user interface...
--:-- Paul Graham: Because you thought we can make the 60%, why not make the rest?
--:-- Amjad Masad: Well, kind of. I think the thing that was very enlightening is watching some of our users code. Some users were doing 100% of their code using AI.
--:-- Paul Graham: No, you're talking about once you released Vibe coding.
--:-- Amjad Masad: No, no, before that. Before that, I'll tell you.
--:-- Paul Graham: How are they finished?
--:-- Amjad Masad: So they would open ChatGPT in one window, and they would open Replit in another window. This is someone who's never coded before.
--:-- Paul Graham: How could you tell that they were pasting code out of ChatGPT?
--:-- Amjad Masad: I invited some users to the office.
--:-- Paul Graham: You would literally watch them?
--:-- Amjad Masad: We would watch them. And basically they would copy the code, put it into Replit. Replit would produce some error. They would copy the error, put it into ChatGPT. And then copy the fix, put it in. It was such a manual process. I looked at it and thought, all we have to do is make an API. Well, we have to automate the entire process.
--:-- Paul Graham: It wasn't obvious. It wasn't easy because back then, AI models would do a couple of tool calls. What's different? The difference is it's not just you have to automate the back and forth, you have to understand the errors.
--:-- Amjad Masad: Right. Exactly.
--:-- Paul Graham: Yeah, that's hard.
--:-- Amjad Masad: Yeah, and you have to loop. And the terminal condition should be that the program is working.
--:-- Paul Graham: Yeah, that's still true. As opposed to kind of like, we'll do one turn, because that's how ChatGPT sort of worked. That's how most AI tools worked. And so can you loop indefinitely? Go to completion.
--:-- Paul Graham: That's the new thing. That's what Claude does.
--:-- Amjad Masad: Yes. So that's what we demonstrated. That's what Replit first demonstrated. We demonstrated a couple of things. One is that you can run LLMs to completion.
--:-- Paul Graham: Which I bet was news to the model company.
--:-- Amjad Masad: They did. They actually reached out to us and they were like, we're so surprised that Replit Agent works. How did you make it work? And the crazy thing, Paul, is that the harness in 2024 is 10 times more complex than the harness today, because the models were less capable. So as models get more capable, the engineering you have to do to make them become agents is less and less.
--:-- Paul Graham: Does that mean it's easier for people to compete with you?
--:-- Amjad Masad: Yes. But we have to do...
--:-- Paul Graham: Well, that's the thing about the AI business today. You have to just accept that everything you're doing is getting commoditized constantly. And you have to be doing the next thing.
--:-- Amjad Masad: Be doing the next thing, right? But the other big thing that we demonstrated is that AI can provision infrastructure.
--:-- Paul Graham: And you actually had infrastructure to provision. That's the thing that's different between you and all these competitors.
--:-- Amjad Masad: Exactly. So we had built databases...
--:-- Paul Graham: You'd spent all these years suffering doing system software.
--:-- Amjad Masad: Right. So it was a little like tool call, spin up a database, spin up a Redis, host this. And so what people found so surprising when we launched... I mean, you saw the product July 2024.
--:-- Paul Graham: And when did you launch it?
--:-- Amjad Masad: September 2024.
--:-- Paul Graham: Okay. And so when people saw it, like Andrej Karpathy said, this was like an AGI moment, a field AGI moment, because everyone was like, holy crap, LLMs could do that. It freaked me out.
--:-- Paul Graham: I didn't like it at all. How do you feel about it right now? Because aesthetically, it was very offensive for you.
--:-- Amjad Masad: Yeah, I know. Because the way I program is the exact opposite of this. I program, I spend three years writing a thousand lines of code. And if I can get it down to 980 in the next month, so much the better.
--:-- Paul Graham: But you gave me the principle for all of this. When you first invested in Replit, when you first backed it, you said there's a super linear relationship between how easy it is to program and how many programmers you get. That's not just true for programming. That's true for anything consumers use. If you make it 10% easier, you don't just get 10% more people. You get twice as many.
--:-- Paul Graham: Yeah, I remember that. I didn't mean make it this easy. You went too far. I mean, just talk in English and it spits out all this code.
--:-- Amjad Masad: Well, the realization I had is for Replit to be truly successful, we needed to go the whole way.
--:-- Paul Graham: Because you did know you were one of the early people predicting AI.
--:-- Paul Graham: And were you predicting vibe coding in something like 2015 or 2017?
--:-- Amjad Masad: Yeah. 2015, 2016, in our original deck, we talk about how if you build a platform where there's millions of people coding, you can train a machine learning model that predicts.
--:-- Paul Graham: That's crazy ahead of its time. This was before the Transformers paper.
--:-- Amjad Masad: That was before Transformers.
--:-- Paul Graham: I'm starting to think that's not visionary. That's just your nuts. If you're that early.
--:-- Amjad Masad: Well, I think I've been fairly good at predicting where the future is headed. Even when I was a kid, I remember when I used my first Nokia phone, I was like, oh my God, this is the new computer. And it was literally a flip phone essentially. Some of that experience also came to Replit. Replit was the first programming platform that had a mobile app. And everyone laughed at us.
--:-- Paul Graham: The idea of programming on mobile. Programming on your phone.
--:-- Amjad Masad: Yes. I had an RFS for it.
--:-- Paul Graham: Well, you told me you would fix bugs using the Hacker News list prop from your phone.
--:-- Amjad Masad: That's right, yeah. Not voluntarily. It was just the only way we could get internet access. But since this podcast is about building the future, I think that people underestimate their ability to potentially predict where the future is headed and build towards it. You just have to sit down and think. I don't think many people sit down and think.
--:-- Paul Graham: This is what experts think. Experts think, all I do is I just sit down and think and I have the ideas. And everybody else tries sitting down and thinking and they don't have the... What's the secret?
--:-- Amjad Masad: It's because you've been doing this idea for like 20 years.
--:-- Paul Graham: When did you first start thinking about making something like Replit where you could program online?
--:-- Amjad Masad: Well, it was when I had my Symbian smartphone and I was in class and I was like, I wish I could program.
--:-- Paul Graham: So how long ago was this?
--:-- Amjad Masad: This was like 2008.
--:-- Paul Graham: Okay, so almost 20 years ago. This is why when you just sit down and think you come up with the future. Because your whole brain has been thinking about this for 20 years. Obsessively.
--:-- Amjad Masad: Yes. You were trying to build Replit before it was called Replit for years and years. This has always been the big thing.
--:-- Paul Graham: So of course you thought of stuff before everybody else. So you knew back to that famous demo in the summer of 2024. When you called me a Bond villain.
--:-- Amjad Masad: Well, you seemed like a Bond villain.
--:-- Paul Graham: That's what I felt like. That I was being shown this frightening thing. I remember what you said: "Don't look at that. Don't look at the code. That's object code."
--:-- Paul Graham: And I'm like, oh no, the stuff I make, the whole stuff I've spent decades making, is just object code. And now everyone's realizing that.
--:-- Amjad Masad: I've been seeing engineers that were also repulsed by that. I remember your tweet was very controversial when you tweeted that thing about the guy who generated vast amounts of code using AI.
--:-- Paul Graham: Even that tweet about me being the Bond villain and object code was also controversial. But now all engineers are... I think with AI, everyone will have to go through this period of denial, then later on depression, and then eventually acceptance. I've had it in like 2020.
--:-- Amjad Masad: Really?
--:-- Paul Graham: And so it just makes it easier for me. I had the grief of the craft that I love is gone. I've already grieved.
--:-- Amjad Masad: You felt that too?
--:-- Amjad Masad: Yeah, yeah, yeah. So you knew what I was feeling when you showed me that demo. But you had to feel it so that you can let go. I think everybody got convinced at the end of last year, at the end of 2025, is when everybody realized, OK, AI is going to write all the code.
--:-- Paul Graham: It's always surprising to me what is the trigger for these kinds of things. Because again, it seemed obvious to me.
--:-- Amjad Masad: Well, for Replit users, I think they already knew.
--:-- Paul Graham: For Replit users. But the thing that convinced the rest of the world might have been Claude Code, right? Don't you think?
--:-- Amjad Masad: I think OpenClaw.
--:-- Amjad Masad: I think Claude Code partly, but OpenClaw, I think, was the first time people experienced this idea of LLMs running to completion, where you give it a task and it's able to go all the way until it's...
--:-- Paul Graham: Which you had been doing for a year and a half at that point.
--:-- Amjad Masad: Yes. But I think the form factor of OpenClaw as well, being like you could text it and all of that. I think I sort of underestimate the power of a form factor, especially for everyday people.
--:-- Paul Graham: Did it influence what you've done with Replit? Did you learn anything from it?
--:-- Amjad Masad: Not yet. I've just learned that lesson. I think it will influence things in the future, which is I just assume that I go out and put out a vision and build a product and everyone will be like, oh, yeah, that's obvious. But I think in reality, you need to tinker a lot with how people experience it.
--:-- Paul Graham: Do you still watch users using Replit and sort of see what they're doing and are surprised by it?
--:-- Amjad Masad: Yes, I actually didn't like to do that. But Haya kind of forced me to do it early on. And I would be kicking and screaming because I would feel intense pain when users are confused by the software or hit a bug. I would be hesitant. I would actually be physically...
--:-- Paul Graham: It's mortifying to watch users use your stuff. It really is. But you know why it's mortifying? Because till now you were in denial about the horrors that were being perpetrated daily out there.
--:-- Amjad Masad: Right. And so our company culture became a lot about being very close to our users, primarily thanks to Haya, having that mentality of you just need to constantly talk to them.
--:-- Paul Graham: She also made you very cool looking. That was one of the really unusual things about Replit early on. Amazing, mind-bending graphic design. Way better than some nerdy program online. A company really needs to have that. And it helped a lot with our initial audience, which were a lot of teenage hackers. I thought it was really cool. It made Replit different.
--:-- Amjad Masad: Replit still has something very distinctive about it.
--:-- Paul Graham: It has a very strong sort of character of itself. Which sounds meaningless, but you know what I mean. It's a very strong solution, not a weak solution. Replit is a strong solution.
--:-- Amjad Masad: Our conceptual understanding of what we build and our product is we always think about engines and hard mechanical things.
--:-- Paul Graham: That's a lot of what that graphic design was about.
--:-- Amjad Masad: Yes. It's like there's something that's sort of hardcore about it, but it's also paradoxical. We slave away so much to make the UX so simple for users. But on the back end, really ugly engineering is happening. Systems programming. Real systems programming.
--:-- Paul Graham: That's part of your DNA.
--:-- Amjad Masad: That's part of our DNA. But it's been challenging because a lot of programmers will look at Replit and dismiss it because it looks very simple. But then even people who end up coming to work at Replit are like, oh my God, there's so much stuff behind the scenes.
--:-- Paul Graham: You do system administration for people's sins.
--:-- Amjad Masad: We do it so they don't have to do it.
--:-- Paul Graham: Do you think that will be the thing that helps you prevail over competitors? Because that is one difference. The depth of stuff that you've built and the depth of understanding you have of how to deploy programs online.
--:-- Amjad Masad: That is helping us win enterprise in our category because they are a lot more sensitive to reliability, security, all of that stuff. It's really hard to get consumers to care about security. A lot of our competitors do really reckless things with regards to building apps that are unsecure. Because if you have agents writing all this code, who knows what it's doing?
--:-- Amjad Masad: And so we put a lot of guardrails, which makes our products sometimes slower, sometimes more expensive. But enterprises really like that. And so these newly arrived opportunists don't have the same depth of infrastructure. Replit's like this iceberg where there's this visible bit above the water that's like, look, you just tell it in English what to do. And it writes your code. And then underneath, there's all this stuff for deployment and sandboxing. We have kernel operating system engineers. We built our own file system from scratch to be transactional and fully reversible.
--:-- Paul Graham: I remember hearing all these stories back in the days before Vibe Coding. That's what all the updates from you would be about. We made our whole new file system. I hope somebody appreciates it.
--:-- Amjad Masad: When you want to make coding collaborative, you had to invent a lot of core primitives.
--:-- Paul Graham: So that's an interesting question. Did the stuff that you did in version one of Replit, where it's just online programming and it's collaborative, did the stuff that you did to make it collaborative also help with Vibe coding?
--:-- Amjad Masad: 100%. It's surreal how plug and play it was. So when you put an agent into a REPL, as we call it, it's like you're collaborating with the agent. The agent is another user. So the initial version was basically the agent was just another user on the same system. And you had everything else built, just ready for it.
--:-- Paul Graham: And even you and the agent could work on the same project and it would just work. So it's no wonder you discovered Vibe Coding first. You were perfectly positioned. You had all the other ingredients. This surreal thing, even the way the company was organized was so perfect. So we had the infrastructure team that handles code execution, the sandboxing team. And then we had the deployment and cloud services team. You can spin up databases and deploy VMs. And then we had the team that was building the glue between the infrastructure and the IDE. And then we had the IDE team. And so where does AI fit into all of that?
--:-- Amjad Masad: We ended up creating a new agent team and we put it in the middle. And all these people are building now for the agent. So the infrastructure team is exposing tools for the agents. The IDE team, instead of building IDE for humans, now they're building edit tools for the agents.
--:-- Paul Graham: Really? What sort of IDEs do agents even need? I thought they just generated code.
--:-- Amjad Masad: Well, initially in 2024, editing code was actually a hard problem.
--:-- Paul Graham: You mean having it go back and fix mistakes?
--:-- Amjad Masad: Yeah, so there was no native edit tool. Now there's a native edit tool every foundation model gives you. But back then we had to create our own edit tool. When you give it a chunk of text, where to replace, you couldn't rely on line numbers because they would hallucinate line numbers. There were all of these different things. You had to make it ridiculously robust in the early days.
--:-- Paul Graham: The LLMs were so bad.
--:-- Amjad Masad: They would also be lazy. They would write a function and get lazy in the middle, like "TODO: finish it later," which is not executable.
--:-- Paul Graham: That was insane. And that was both OpenAI and Anthropic had the same bug. I think it's because of the chat model. They always wanted the chat to halt early, not produce mountains of text. So that was an artifact of that. And so we had to detect that and loop over.
--:-- Amjad Masad: So even editing text was a lot of work. And now there's a native edit tool.
--:-- Paul Graham: Is there anything like an IDE for agents though?
--:-- Amjad Masad: In the initial days, there was a lot of research on...
--:-- Paul Graham: No, I mean now, the way agents work now.
--:-- Amjad Masad: Now you can just give them a shell. You just give them a shell and they'll write shell commands. They can do everything with cat and basic shell commands. They're very good at shell.
--:-- Paul Graham: You needed to have the IDE early on because the agents were bad.
--:-- Paul Graham: And so that was one of the things that enabled you to be so early.
--:-- Paul Graham: Could you have done it any earlier?
--:-- Amjad Masad: No. Because you got it just at the point where it barely worked at all.
--:-- Paul Graham: And it barely worked. When we launched it, it was in beta. I would say 50% of it... I remember I would go over to your house and you would show me things and they would not work.
--:-- Amjad Masad: Yeah, maybe 50% is too harsh. I would say 30% of the time it would not work.
--:-- Paul Graham: So why would users tolerate that?
--:-- Amjad Masad: Because it was such a magical experience. And also, when they were cutting and pasting stuff back and forth between ChatGPT, that also wouldn't work either. But if they couldn't program at all, then programming 70% of the time was a huge improvement. The initial set of customers that started paying us a lot of money were people who would otherwise go on outsourcing sites and get outsourced programmers. So now you either pay $100,000 to get an app done by an engineer somewhere, or you pay Replit $1,000. It was a two-order magnitude difference. It still is today between getting an engineer and having Replit do it.
--:-- Paul Graham: Arguably the whole existence of the outsourcing shops was evidence of a market there. Because I thought of them as just terrible. Who would ever want this? Who would pay money to these people to make vast amounts of terrible software? But instead I should have thought, wow, that's a sign of demand.
--:-- Amjad Masad: We try to satisfy this demand in a few ways. Initially when we had all these teenage hackers on the platform, we built something called Replit Bounties. I don't know if you remember that.
--:-- Paul Graham: I remember Replit Bounties.
--:-- Amjad Masad: It was trying to get at that. I was like, okay, AI is not there yet. They were gonna become the new outsourcing shops, just piecemeal, a bunch of teenagers.
--:-- Paul Graham: But the insight there was that they were powered by the rudimentary AI we had on Replit back then. That in itself was enough of an edge.
--:-- Amjad Masad: The people doing the Replit bounties were already using AI to write the stuff.
--:-- Paul Graham: Exactly. And it grew really fast. But then I looked at it and I was like, okay, if this is where the future is at, it's a dead business. So we shut it.
--:-- Amjad Masad: Why would it be dead?
--:-- Paul Graham: Because ultimately people were just asking agents for the entire thing. Not even going through the teenage kids asking the agents on their behalf.
--:-- Amjad Masad: That's the case right now. We go into big multinational corporations, banks, things like that. They have a lot of SaaS software they pay $10 million for, or old software. And they find a few vibe coders inside the company that have adopted Replit, and they're going and replacing all these different systems. It starts with internal stuff.
--:-- Paul Graham: It starts with internal stuff.
--:-- Amjad Masad: Previously they would tell us that they scoped this project out and it would have been $10 million. We spent $250,000 on Replit. So it's happening all over the place. I think the outsourcing industry is sort of dead at this point.
--:-- Paul Graham: So is it mostly not very technical companies that you're talking to? Companies that don't specialize in technology?
--:-- Amjad Masad: We have a lot of tech companies as customers. I think that's not where we have the biggest impact. Because I think they can fish for their own. But the banks, the manufacturing or ordering companies, they need something like Replit. Otherwise, they'll shoot themselves in the foot. It's driven mostly by people internally who are ambitious, like some guy who's in charge of sales and wants to write some tool for his sales department.
--:-- Paul Graham: This created a bit of an identity question inside Replit. It was like, the enterprise, that's not part of our mission.
--:-- Amjad Masad: But what I tell my team is that those are entrepreneurial people. Those are ambitious people building businesses. This feels very core to what we do.
--:-- Paul Graham: Those are scrappy people within companies too.
--:-- Amjad Masad: Exactly. And even if the companies aren't scrappy, the people are. And they tend to get promoted. A lot of people who bring Replit into their organization will get promoted because they are having great impact. Especially now, all these companies are thinking, we got to use AI. I keep reading how AI has taken over the world. We ought to be using AI within our company. Who within our company understands anything about AI? Who's using it? And here's the person using Replit to make stuff.
--:-- Paul Graham: They're going to be the ones that they trust.
--:-- Amjad Masad: Exactly. So it's a continuation of that, but there's a bit of schlep blindness. There's a lot of boring work to be enterprise ready.
--:-- Paul Graham: You have practice doing that kind of stuff. It's just like systems programming.
--:-- Amjad Masad: A lot of it is also just giving admins and IT a lot of controls over what people could do and setting budgets.
--:-- Paul Graham: They want to have rules.
--:-- Amjad Masad: Yeah. It's sort of like sandboxing for humans. It's the principal actor problem. The end users are not actually decision makers for the buyer. That's why most of the history of enterprise software has sucked. But now I think people have a lot more say on what they want to use, and therefore products that are really good are winning in the enterprise. But that problem still exists. Sometimes IT and security will force you to make the product worse for the users.
--:-- Paul Graham: Do you think the answer is give them what they want? They want to do stupid things, let them do the stupid things. Or do you think the answer is secretly you're working on behalf of the ambitious, scrappy people in the company and you secretly want to give them a way to get around the bureaucrats?
--:-- Amjad Masad: That's a tough one. For example, I disagree with rationing tokens. The way companies are approaching this is saying everyone gets $100 of tokens a month. What they should care about is ROI. And what we see from our customers is there's a power law of ROI inside companies. One person could drive 100x the next person.
--:-- Paul Graham: I bet there is. Absolutely. There always is in software. And especially with agents, it used to be 10x, now it's 100x, it's 1000x.
--:-- Amjad Masad: You want to control token spend. So what you need is to help the companies see where good stuff is happening. I think there is some invention we could do there to help with that. In the short term, you kind of have to do what they're asking for. That's how you get in. And then you want to invent something better and then show them the way.
--:-- Paul Graham: So there's more stuff to build.
--:-- Amjad Masad: There's more stuff to build.
--:-- Paul Graham: This could be very valuable stuff. When you think about what you're doing when you're trying to figure out which of the projects that people are using tokens for are the most valuable, you're fixing the biggest problem inside big companies. This is why big companies are the way they are. They can't figure out the value of stuff done by individual people. So if you build something that makes them even 10% better at that, you're fixing the whole fundamental problem with big companies.
--:-- Amjad Masad: I don't think we describe ourselves as vibe coding now. We wrote this blog post called "The Self-Driving Company." The idea behind it is at Replit, we're early adopters of technology, probably, but definitely AI. We're seeing that we're working differently. There's a fundamental shift in how we work.
--:-- Paul Graham: You mean Replit as a company?
--:-- Amjad Masad: Replit as a company, which is customer zero of Replit.
--:-- Paul Graham: So maybe you should figure this out yourself first. How do you identify people doing valuable things?
--:-- Amjad Masad: Yes. And so Replit as a customer zero is the lab. We're trying to figure that out. The self-driving company is this idea of the future of work, future of how firms are organized. And we think Replit is the platform for the transformation for self-driving companies. That's the expanded vision of where Replit has had it. Because ultimately, the thing that I didn't really appreciate and realize is, I've always known that making software creation more accessible would have transformational impact. I didn't really appreciate that it would fundamentally change how people work.
--:-- Paul Graham: Because we were programmers. It seems like programmers write programs. It's not like the core of the way companies work consists of programming.
--:-- Amjad Masad: But maybe now it does.
--:-- Paul Graham: It does, yes. Wow. By the way, incidentally, this is an instance of what happens with the very best startups. You come up with something and you think you're coming up with some idea you suggested to them. They've already done it. You've already got this big new transformation plan. Because that's another thing about the best startups. The best startups, they're already big. And then they think of this new thing as if they were a brand new company. And they get 100x bigger.
--:-- Paul Graham: Well, I think it's now actually essential to do that because AI, like we talked about earlier, is a massive commoditization force. So I think where it used to be optional to have to reinvent yourself, it's now mandatory almost.
--:-- Amjad Masad: And you're the way AI flows into companies.
--:-- Paul Graham: Yes. I mean, that's gigantic.
--:-- Amjad Masad: Yeah. That's gigantic.
--:-- Paul Graham: Yeah. You just have to make that be not vague because anybody can say that. Anybody listening to this, there's probably someone at IBM thinking, yeah, that sounds good. We'll say we're the way AI flows into companies.
--:-- Amjad Masad: Everything we've kind of popularized ended up being copied. Like, we popularized this idea of "idea to app fast." And now every startup that does, like, five pudding is like, that's their technology. But they're copying the old version of you.
--:-- Paul Graham: Yes. You're a movie, and they're copying this frame from six months ago, right?
--:-- Amjad Masad: Yeah. That's how it needs to be. You need to be constantly doing that.
--:-- Paul Graham: But this means Replit is in every company. Every company that has competitors can't afford to just sit.
--:-- Amjad Masad: Yes, that's the goal.
--:-- Paul Graham: That's astounding. So how do you do this? I suppose gradually is the answer.
--:-- Amjad Masad: Yeah, I think there's always this temptation for founders to come up with this big strategy, make a big news about it, and just say, "Oh, this is what we're doing now." It's like, no, let the customers tell the story. You have this general goal of where you want to end up, but then you do it incrementally.
--:-- Paul Graham: And what are the increments? Individual things that come up with customers, but you know what you're looking for. You know you're looking for a customer doing a certain sort of thing. Okay, let's help them out and then everybody will be able to do this.
--:-- Amjad Masad: Exactly. And that's the great thing. This is what having done all that systems programming buys you. It gets you into the customers so you can see the increments of the new things you're doing.
--:-- Amjad Masad: And similarly, how Replit Agent for the outside world was this discontinuous moment, for us internally it was incremental in many ways. I think a lot of things that we're doing, at some point we were like, there's this discontinuous moment in how companies operate. It's so interesting. If you're the experts, if you're the insiders, things that seem like step functions for the rest of the world, you saw before the step and you see after. So it's smoother for you.
--:-- Paul Graham: It's smoother. And I think that not all invention, but I think the history of invention, if you zoom in and read the stories, Steve Jobs tried to do an iPad in the 80s.
--:-- Amjad Masad: Yeah. Because he knew that something like that would be good.
--:-- Paul Graham: Yeah. The technology just wasn't there yet. And so when you read the story of Apple, it was like, oh, the iPhone was totally obvious. Especially after it.
--:-- Amjad Masad: Yeah, yeah.
--:-- Paul Graham: If it was so obvious, how come nobody else did it? Right. But it was sort of incremental inside Apple, but for everyone in the world, that was a totally transformational moment.
--:-- Amjad Masad: But also, if you know the story of how the iPad happened, or the iPhone, originally they were making this tablet and he already knew what he wanted. He'd been thinking about this. And so as soon as he saw it, like, okay, this is it. Let's make this.
--:-- Paul Graham: And you too. Whenever something happens within a company, it's going to be something that you've been thinking about for 10 years and you're like, okay, this is the thing to build. More than other companies though, I think you guys have been living in the future. Don't you think that's true?
--:-- Amjad Masad: I think that's true, yes. That's almost, if you had to say what's the real DNA of Replit, it is living in the future, for sure. More even than AI specifically. It's often frustrating.
--:-- Paul Graham: Because stuff doesn't work.
--:-- Amjad Masad: Well, stuff doesn't work and people change so slowly. So for example, cloud development. Are people convinced about that now? Not yet. Still, how are they going to use AI? It's crazy, exactly. So our prediction internally is that you go from everyone using one agent to 10 agents, to a hundred agents, to 10,000 agents at some point. You can't run 10,000 agents on your MacBook. There's no way you can do it.
--:-- Paul Graham: Yeah, Robert proved that back in the 1980s.
--:-- Amjad Masad: Yes. So eventually we're going to get away from local work in general, but people still predominantly do work on their desktop. Maybe this is what will finally convince them.
--:-- Paul Graham: Yes. It still has to live in the cloud.
--:-- Amjad Masad: Yeah, but it's also fun. I think working at Replit just ends up being a weird sort of existence. You go into this company and you're like, oh my God, I'm seeing...
--:-- Paul Graham: Well, the weird thing is you know what everyone's future is going to be like. Because that was why you seemed like a Bond villain when you were demonstrating what later came to be called vibe coding. And plus that laugh, right? You were demonstrating it and here I was sitting there looking at it with dismay. And you're like, "Yes, Mr. Bond. Accept it now."
--:-- Amjad Masad: Exactly. This is the future, Mr. Bond.
--:-- Paul Graham: I think you might even have been wearing a black turtleneck.
--:-- Amjad Masad: I was. Not turtleneck. It was a black shirt.
--:-- Paul Graham: Okay. Yeah. The black shirt, the bald head, the beard. Me sitting there like, oh, no, what's happened to my beloved programming? And you're like, "I already knew. I've been thinking about this for years, Mr. Bond."
--:-- Amjad Masad: I do think I have an ability to kill the sacred cow in a way. When computers got really good at something, I'm like, all right, it's done. Let's move to the next thing. I'm not so tied to what I do. I'm fine with throwing away things and starting over. In my career, I've started over many times.
--:-- Paul Graham: I remember you were always rewriting things within Replit.
--:-- Amjad Masad: Yeah, rewriting things. But I think most people, they're just kind of rigid about, this is my identity, this is what I do.
--:-- Paul Graham: I mean, they should read "Keep Your Identity Small," one of your essays. But I think this moment requires a ton of adaptability. A lot of people ask me, how do you raise children and what do you teach them? And it feels like the main thing you teach them about AI is flexibility and adaptability and being a lifelong learner.
--:-- Amjad Masad: What I tell them is that if you have this tidal wave approaching the beach, the safest place to be is surfing on it, not standing on the beach holding onto a palm tree. It'll just get washed away. And that's what your users are doing. To the extent anybody can see the future and be protected against bad things happening, your users are it because they're all using AI. If you want to use AI, Replit is the way to use it. And if you're using it, it's the best way not to be squashed by it.
--:-- Paul Graham: Right. And this is why we feel our mission has a moral angle to it almost. You want to help these companies transform. Get those people off the beach and get them all onto surfboards.
--:-- Amjad Masad: That's right, exactly. We did this hackathon with a very large telecommunication company. They did a survey before they used Replit and a survey after they used Replit. The survey asked them how anxious they were about AI, how worried. Pre-using Replit, just for an hour, 80% were anxious and worried about their jobs. After they used Replit, it flipped. 80% were optimistic.
--:-- Paul Graham: For one hour.
--:-- Amjad Masad: One hour, yeah. And I think people in Silicon Valley constantly underestimate how the rest of the world have not experienced what we've experienced just yet. And it's just going to take a long time because things are moving so fast.
--:-- Paul Graham: Well, no, but that's what I'm saying. If they're just two years behind, that's forever behind.
--:-- Paul Graham: Because two years ago, vibe coding didn't exist.
--:-- Amjad Masad: Yeah. I mean, most banks are using ChatGPT like 2023, basically. The thing that is rolled out at most banks, other than engineering, is ChatGPT 3 or something like that. It's just an LLM barely connected to anything. So it's not that useful. There's a real lag for what people use. Silicon Valley already went to these agents that are doing things for you. But most companies are still trying to figure out what to do with all the legacy software that they have that they want to rewrite. And they can't just ignore it.
--:-- Paul Graham: Maybe all this COBOL code will finally get rewritten.
--:-- Amjad Masad: It is starting to get rewritten.
--:-- Paul Graham: Really? The COBOL is finally going away.
--:-- Amjad Masad: Yeah, yeah, yeah. COBOL would eat up a lot of tokens. Actually, early on in our AI journey, a lot of companies would be like, "Oh, this Replit thing is cool, but why don't you come rewrite our COBOL?" I was like, "We're not going to write your code for you. We'll teach you how to use this thing." But now there's a bunch of AI companies that will come and rewrite your COBOL.
--:-- Paul Graham: So there is one last question I want to ask you. This whole Replit, for its entire history, the reason it has worked is because you're living in the future to such an extent. You personally are living in the future. How do you do that? Why are you living in the future?
--:-- Amjad Masad: I sort of started answering some version of that early on in our conversation. And you had a good point that it's not just about sitting and thinking. But I think a component of it is that I actively try to predict things and I actively revise my predictions and look at them. I think it's a skill you can get good at. And there are some certain principles that you can follow. Early on, I read Ray Kurzweil. His whole thing is if you chart a log-linear chart of compute flops in the world, it's this straight line. And actually, AI didn't bend it, which is crazy. It's still a straight line. It needed AI to happen.
--:-- Paul Graham: It needed AI to happen.
--:-- Amjad Masad: It's almost a religious sort of thing. It's like it'll happen anyways. I saw someone tweet, he said, "It's not that AI brought on the compute, the compute brought on AI." That's true. Which is a fascinating thing. Deep learning would not have happened without the compute. Everybody thought neural networks didn't work. Back in the 80s, that was the thing about neural networks. They didn't work. But it turns out if you give them 10 billion times as much resources, then they start to work.
--:-- Paul Graham: Yeah. Who knew?
--:-- Amjad Masad: So that line, there's a lot of predictions you can make by truly understanding a line.
--:-- Paul Graham: More broadly, the reason you're so good at living in the future is because you have normalized that line. You assume that you're not living in a static world. In fact, you're living in an exponential world. And a lot of what you talk about in your essays is understanding the exponential. I think once you truly understand the exponential, you can look at an LLM. I can look at GPT-2 and extrapolate a lot better than most people.
--:-- Amjad Masad: I imagine chats, I imagine agents, I imagine personal assistant, all of that back in 2019 when I first saw GPT-2. But you just have to believe. Most people have this suspension of disbelief. You have to actually believe, you have to truly embrace it and go against your instinct, because I think there's a human instinct...
--:-- Paul Graham: Oh, there's all these reasons why this would not work and all these problems. Well, people implicitly treat things as static. And the difference with you is that your default assumption is that things will change exponentially.
--:-- Paul Graham: That's what's unusual about you.
--:-- Amjad Masad: Yes. So having embraced that as a kid, reading sci-fi and reading Kurzweil and all of that, I think it created two things in me. One is an ability to accept change and be okay with change. Computers are going to do more and more of the things of my skills.
--:-- Paul Graham: Not just accept, expect. Expect and then build towards that and skate where the puck is going. And the puck is moving faster and faster.
--:-- Amjad Masad: It is moving faster and faster. That's what it means. But I feel more in my element right now.
--:-- Paul Graham: Because nobody else can keep up with you. That's why someone who assumes dramatic change, imagine who would be perfect for this world where everything is changing under our feet every six months. The person who assumes dramatic change. You're just perfectly suited to this moment. That's why Replit's doing so well.
--:-- Amjad Masad: Thank you.
