Amjad Masad of Replit: 10x’ing in a Year and Building the Future of Code | Frameworks for Growth

🔔 Subscribe for more startup strategy and founder stories: https://www.youtube.com/@TrustVanta

👉 Follow us on all social media platforms: LinkedIn, X, Instagram, and TikTok

🎧 Follow Frameworks for Growth on Spotify and Apple Podcasts

‍

In this episode

Vanta CEO Christina Cacioppo talks with Amjad Masad, Founder and CEO of Replit, about how persistent ideas evolve into breakout products and how founders can stay scrappy while scaling.
‍

Amjad breaks down how Replit handled early competition, carved out space as one of the first AI-native dev platforms, and sustained momentum in a crowded, fast-moving market.

‍

🔑 Topics covered (timestamps):

00:20 - Introduction

00:41 - Software development trends Amjad is excited about

02:28 - What is Replit

03:21 - How long Amjad has been working on Replit

04:53 - How did Amjad start learning Javascript

06:34 - Why the web felt like the future

08:47 - Why AI feels like the future

12:15 - When models will solve coding

14:24 - How Amjad practices what he does

16:05 - Have the things that cause pain changed

17:22 - How Replit compounded

19:22 - Why Amjad learned SEO

20:12 - The day Replit went viral

24:18 - Navigating skepticism

29:18 - Hiring high agency individuals

35:25 - Replit's innovations being coped by giants

42:36 - Why Amjad wrote an AI manifesto

49:36 - Replit's approach to marketing

52:44 - Amjad's advice for X/Twitter

55:03 - Amjad's thoughts on building community

56:34 - Why powerlifting is similar to building a startup

57:25 - Reading philosophy and influences on AI

1:00:07 - What Amjad would be doing if not working on technology

1:01:43 - Why Amjad chose Foster City

1:03:34 - How Amjad would spend a day with a founder in Foster City

1:04:51 - Unconventional advice for founders

1:08:00 - What Amjad is reading

1:09:34 - The first program Amjad ever wrote

1:09:54 - Amjad's thoughts on layoffs

1:11:26 - Choosing between making Replit bespoke or general

‍

Read the full transcript

Christina:

Welcome to Frameworks for Growth. I'm Christina Cacioppo, CEO and co-founder of Vanta, and I'm here today with Amjad Masad, CEO and co-founder of Replit. So prior to Replit, you built big parts of the JavaScript infrastructure at Facebook. We were batchmates in YC in Winter 18 and have known each other long before then. Thanks so much for joining me today.
‍

Amjad:

Thanks for having me.
‍

Christina:

Amjad, you have been thinking about software development and how to make it easier for almost two decades now.
‍

Amjad:

Yes.
‍

Christina:

What trends are you really excited about now that most of us don't know about yet?
‍

Amjad:

Don’t about yet. I don’t know if people don't know about yet, but I don't think people understand how fast it's coming, how fast software agents are going to get better and better. Right now we have this version two of agents in beta. We're looking at the metrics. It's something like five times better and by the way, version one was released in September.
‍

Christina:

Oh like six months.
‍

Amjad:

Shorter. Yeah, so I think just this year large parts of software engineering is going to be automated. The use cases that we're focused on will especially have tremendous gains because when you're trying to add a software development agent to a big company repository, it's hard for it to make an impact. But with Replit, people are making applications from scratch, so I think agents in general are going to start to work this year. We have reasoning models. It's going to be a really exciting year.
‍

Christina:

Neat. Are you all using agents to build Replit?
‍

Amjad:

Yeah, a little bit. Definitely, we're using Replit agents internally. One of my favorite use cases here is that one of our HR team members, her name is Kelsey, she was looking to replace an org chart software we're using. They all kind of sucked in their own ways.
‍

Christina:

Yes.
‍

Amjad:

So she decided to build her own and three days later she had her own, she had all the features she liked and so not only we are building our software and software engineering team, but agents and Replit agents is making everyone more productive.
‍

Christina:

And what is Replit for folks watching?
‍

Amjad:

Good question. So maybe a year ago my answer would be different, but right now it's totally transformed. Basically it's a way to build software entirely using natural language, so you can just talk to your phone or computer and it will write the code for you. It will set up all the packages and libraries. It will provision cloud databases and storage and all of that and then you can iterate with it. You can change the design, you can add features, you would talk to it as if you're talking to a human engineer and then when you're ready you can deploy it and all of that happens on this one app. We spent the last seven, eight years building all these components and we're the only product on the market that has all these components.
‍

Christina:

You mentioned building for the last seven, eight years, but if I have this right, you've been working on Replit for longer. How long have you been working on Replit or something Replit-shaped?
‍

Amjad:

I think the first Replit-shaped idea was probably 2008, 2009. I was obsessed with programming. I wanted to try all these different programming languages. I wanted to code my phone. I had this Nokia Symbian phone those really big phones. I was obsessed with it. I was like, oh, I really want to code on, it now we have a coding mobile app. The thing that was really annoying about doing homework or learning a new programming language is all the setup you have to do, all the IT chores you have to do. You have to download an IDE, which sometimes gigabytes big and you have to figure out the packages and god forbid you want to share the program, it almost never works on someone else's machine. So at the time everything was going to the cloud. I remember using Gmail at the time and Google Docs and Chrome had just come out and Chrome was really, really powerful. So it was like, oh, why can't we code on the browser? And I thought that'll probably be a weekend project. Little I know here 20 years later I'm working on it, but I did get something out in a weekend and it was like a small sort of sandboxes. You could type a lot of JavaScript and run it and it was already pretty cool.
‍

Christina:

You probably are one of the people most advanced in the world at about JavaScript. How did you start learning JavaScript and not just alert box JavaScript but Chrome, what you can do in new Chrome or 2008 Chrome?
‍

Amjad:

After I finished college back in Jordan, I went to work for Yahoo. Yahoo had just acquired a company, an email company called Maktoob and it was the first big acquisition in Jordan, maybe even in the Arab region. So I went to work there and Yahoo's very well known sort of web, JavaScript shop, so I got really attracted to the language because it had this sort of scripting angle. It was also functional, it was reading a lot of Paul Graham, his essays about program languages and Lisp and all of that. So it was a very interesting language but I also felt like I can have a lot of impact with it. I also felt like it's like a bet on the future and I think this is something I really try to do and I think I have an eye is where the future of technology is headed and I felt like JavaScript is on this trajectory where you had Chrome optimizing it using V8 and you had these web applications like Gmail and Docs and all of that, and so I started really investing in it. And a lot of it is practice. The reason I built Replit is to practice programming. I think you need to just put in the hours and if it is fun for you, you're just going to be obsessed and going to keep doing it.
‍

Christina:

I have so many questions. What was it about the web where you felt like this'll be the future? Because now it sounds so obvious, but Gmail is kind of weird. People wouldn't totally use it. Google Docs had very low adoption—wasn't foregone conclusions.
‍

Amjad:

I started programming in BASIC and then in Visual Basic, and I built my first sort of business when I was a teenager selling VB6 applications.
‍

Christina:

And who'd you sell those to?
‍

Amjad:

Yeah, so I was obsessed with playing Counterstrike and so I would go to these internet LAN cafes and play Counterstrike in LAN with my friends and I was just surprised by how much manual work was going into managing these stores. They'd have a lot of employees and when you go in you'd pay for an hour, they'll write your name down and they have to watch the clock and go tap you on your shoulder. So what I did is I wrote this client server management software and it had security features so that people can only boot up the games or enter browser and was able to sell it to a bunch of stores in Jordan. Throughout all of that, I just learned how painful it is to make and distribute software, and it's also the story behind Replit. It's like for, I actually spent two years when I was 12, 13, started working on it. I only got it to market when I was 15.
‍

Christina:

Replit you mean?
‍

Amjad:

No, no, no, no, BASIC when I was 15, the Visual Basic kind of application. And throughout that I just learned how just painful it is to distribute software. So when I saw the web, the thing I saw, kind of, web applications working, I was like, oh, this is how you distribute applications. You don't have to go buy CDs, you don't have to move software in this way. You just go to a website and that's your application. You can progressively load it and do all these things and also you see all the attention going into it, all these companies kind of optimizing for it and felt like it was the future.
‍

Christina:

And then what is it about AI now or that some of that, because it seems like some of that same, oh this is the future is going off for you?
‍

Amjad:

Yeah, so actually it goes back to my time at Code Academy. So Code Academy is like you can go to it and type in a bit of programming, it'll tell you whether you're doing good or not, you go to the next exercise. And so it's learn by doing programming, and I was trying to give users feedback about their code and I was doing all this sort of manual regex to just check for completeness, and I just felt like I was building all these tools for code and I just felt like, well, a lot of this work is really manual and deep learning was getting good at natural language processing and I felt, well, can't you apply neural networks on code? And I looked at it at the time and there was some initial progress there, but it wasn't there yet. Then at Facebook I was also working on compilers and things like that and I had that urge to kind of go try to figure it out. I played around with certain technologies, it wasn't there yet. I started Replit the company in 2018, 2019—GPT-2 comes out and I felt like, okay, now is the time, I think you can apply natural language processing on code. We have a lot of code data at Replit. We can perhaps fine tune GPT-2 or train the model, but it was a little too early because the tools were not there. It was really painful to do anything with neural networks and especially transformer models. So it wasn't really until GPT-3 that it was obvious that these models are going to be great at code. Now the next thing you had to sort of imagine because you can really do it very well, is agents. One of the first things I did with GPT-3 was have it produce Python code and then having a Python interpreter run the code and I think I was one of the first people to do that and then Langchain came out and they based the code on my experiments and basically the idea was like, okay, if these language models can call tools and external services, you can see the beginning of what we call an agent today. Of course tool calling is not exactly the only thing you need. It's not sufficient, but if the model can do unbounded set of actions, then I think this is what I define as an agent. Being able to do call tools but do an unbounded set of actions until it naturally find halts, comes to a conclusion. And I would say probably that only became true in last year.
‍

Christina:

Okay. I was going to ask, with a specific model with 4o?
‍

Amjad:

Yeah, Sonnet 3.5, I think was the first time you could see a glimpse of agentic behavior. Then they came out with the next version which had computer use, and you could tell they're starting to optimize for long-horizon reasoning and now 3.7 is very, very impressive and I think other companies are following suit.
‍

Christina:

So if you had to guess when will we solve coding with models? When do you think that'll be?
‍

Amjad:

Like I said, I think for the tasks that people use Replit for, so like MicroSaaS, like, I'm entrepreneur spinning up my startup, I'm building internal tools, I'm building microservices and building AI agents, so not super large code bases, not like 10 gigabytes. I think towards the end of the year it'll feel perhaps like a mid-level engineer. It'll feel like you have a software engineer in your pocket, you'll open Replit on your phone and you'll be able to build an application. I mean we already see this. I was in Qatar the other day. I was in a conference and I was walking around and someone stopped me and I was like, Hey, I work at Qatar Airways and I'm not from the region, I don't speak Arabic. So I was at a meeting where everyone's speaking Arabic, so what I did is I pulled out my phone and I asked it to make a translation app using OpenAI Whisper and whatever. And before they ended the meeting, he had the app. He couldn't use it in that exact meeting, but he was able to use it in subsequent meetings. And so the idea of on-demand software is already here is just going to get a lot more reliable towards the end of the year. Now the question of, when do we get to a point where you can wholesale automate big aspects or large aspects of the software engineering team? I think probably product engineering, frontend engineering, some full stack engineering towards the end of this year as well, we're going to see significant impact on that where teams can be a lot smaller, a lot leaner, where you have AI agents running continuously on your code base and I think maybe by 2027 a single entrepreneur could be spending tens of thousands of dollars on virtual employees.
‍

Christina:

Yeah, very neat. I want to go back to something you said when you were talking about making some of your first, or making the Visual Basic apps and scripts and practicing what you did. How do you practice what you do now?
‍

Amjad:

That's a great question. By writing email, and I just wrote on the keyboard.
‍

Christina:

But how do you practice writing the important work emails, even?
‍

Amjad:

An interesting one is public speaking is what I'm doing here is being able to communicate, sort of, complex ideas. And I think you can do that by progressively putting yourself in more and more difficult situations where you're having to improvise. For example, you can commit to a talk but go there unprepared. Maybe that's the final boss, but you can maybe do it with an interview, with a fireside chat, or something like that first and then just progressively make it so that you actually can think on your feet and come up with ideas in real time. So public speaking is one where it's obvious. We have a value at Replit, it's seek pain.
‍

Christina:

Such a founder's value.
‍

Amjad:

It scares a lot of people from joining the company, which I think is a good thing. And the idea behind it is through pain and uncomfort, growth happens. So I think my version of practicing what I do is leaning into the hard and painful parts of it. If I feel like I'm flinching for something, try to override that and do it even harder and put myself in even more uncomfortable situations.
‍

Christina:

Yep. Has it changed what you find yourself flinching from or taking pain from? Has that changed?
‍

Amjad:

Yes. Like, making large changes to the company. I mean obviously the most painful of them is a layoff, but also just, reorgs, change in priorities. They're all painful, especially you can tell people are disappointed and I've gotten better and better at that and it's not like I don't feel people's emotions, I feel them as strongly, but I'm able to frame it in a way in my mind, that I'm doing the best for the company. Sometimes I'm wrong, but I'm often right and we'll end up net better not just for me, for the employees, for the shareholders, and for everyone.
‍

Christina:

That kind of scale in your mind of immediate pain over here or immediate discomfort or displeasure and then longer term it will be or medium term even, it'll be better and the scale goes that way.
‍

Amjad:

Exactly. I mean it's counterintuitive, but seeking pain is actually net reducing pain, right?
‍

Christina:

Yes. Because it doesn't stay there.
‍

Amjad:

Yeah, well delaying pain compounds it.
‍

Christina:

Yes, it's so true. So speaking of things that compound, we met a long time ago, and there were times when—Replit has been around for how long? Replit the website.
‍

Amjad:

2011.
‍

Christina:

2011. And then there were times when maybe, more or less effort into prior to forming company, but it never, always kind of compounded. Why is that? Most people, when they put something on the internet and then don't pay attention to it, everyone else stops paying attention to it too. That's not the Replit story.
‍

Amjad:

I did quite a bit of work initially on SEO, and I did quite a bit of work on promoting it initially. And those things at least used to compound. Right now the web is totally different. I bought this SEO book, which wasn't that helpful actually, like an actual book. I was really interested and I was like, okay, I've been mostly programming and I haven't done any marketing, but I just learned about getting backlinks from authoritative sites and doing all of that. So I started spamming the web. I got Replit’s link in Wikipedia somewhere.
‍

Christina:

That's probably a really good one actually. High domain authority.
‍

Amjad:

Yeah. And then as it got started, it basically just compounds and people started using it. I'll post the first few links on Stack Overflow, then people started using it there.
‍

Christina:

Like, here's your fix, but go see this Replit.
‍

Amjad:

Yeah. I spent three months answering questions using Replit, and then people started doing it. I'm like, okay, that's done. Let's go to the other one. So you want to bootstrap growth and then that's how exponentials happen, right, you bootstrap the thing and it grows slowly, and then more and more people do it, they tell their friends about it, the links get better over time and so on.
‍

Christina:

How did you know that was something you should be doing? Somewhat, I mean Wikipedia specifically, but just backlinks, SEO, how did you know you should learn about all of that?
‍

Amjad:

I was just attracted to the idea of more people using my stuff. There are two reasons I like that. One is I actually just like the challenge, the infrastructure challenge of scaling things. I want more people to use my stuff because I wanted to break and I wanted to fix it. I wanted to learn how to build distributed systems. And so there's sort of a nerdy reason to grow. But I've always loved this feeling—the first time we went viral.
‍

Christina:

Well, tell me about that. You seem like you remember it.
‍

Amjad:

Oh yeah. It was one of the best days in my life. 2008 all the way to 2011, I'm iterating on this idea. And finally it worked. We had a technical breakthrough and we were the first to run Python on a browser.
‍

Christina:

And how'd you do that?
‍

Amjad:

I had the intuition that what I needed to do to run other program languages is to compile them to JavaScript. I was getting really good at JavaScript and getting really good at compilers, and initially I started writing these languages, these interpreters myself. I'm like, I'm going to write Python and JavaScript, and that was a really bad idea that would've taken me another 10 years to do it. But I found this project that was being incubated by Mozilla Research and what they were trying to do is to take a native program, program written in C, and then compile it using LLVM, Apple's kind of compiler toolchain to something called IR, intermediate representation. So not to machine code. So it's just this intermediate representation and someone in Mozilla figured out that, oh, we can turn that intermediate representation into JavaScript fairly easily. That was the idea—by the way, that idea became Wasm and we have it all over the web, and that guy worked on it. So I found it, it was very, very early and it could barely run something like Python. And the main missing things were creating a virtualized environment on the client side. So this is where we contributed to the project. This is where we spend a lot of our time. It's like how do you make it so that JavaScript or the compiled program thinks there's a file system, thinks it's running on Unix. And so we started writing all these APIs. For example, when it tries to reach for a file, we actually go to the server and then fetch that file and bring it all the way back. So instead of going to disk, it's going to the web. So we got to a point where we're running Python. That was a magical thing. 
‍

Christina:

 I think I remember that, when that happened. When yall did that.
‍

Amjad:

You were trying to build on it, I think, on the open source project, right?
‍

Christina:

Yes, there was this—it was a coffee pun, but it was JavaScript to Java.
‍

Amjad:

Yes, yes. So I tried to launch a few times actually, although we had the second breakthrough, it wasn't really picking up as much as I'd like. I was so surprised. I was like, we had this big thing and almost no one seems to care. So I kept trying to launch, I kept doing it on Twitter, on IRC was still a thing, on Reddit, Hacker News multiple times. Finally, there was one that hit, and I remember the title. It was like, try Python, try Ruby, try Lua all in the browser. And then that got upvoted. It was in Hacker News for like 48 hours and then it started getting picked up on Twitter. Google Plus was still a thing—I'm really dating myself here. And then my friends and I, some of them worked with me on Replit, we were in the desert actually in Petra. We were like, we had been working on this thing for a long time and we kind of wanted to go on a trip and again, we were on these Nokia phones. I mean none of us had iPhones. The Hacker News thread happened and then travel the next day or something like that. And I couldn't enjoy the trip because I kept going to this old Symbian, I dunno what the web used to be called, the web browser used to be called on this and searched Twitter for Replit, just see everyone tweeting about it. And my favorite one was Brenda Ike, the inventor of JavaScript tweeting about Replit. And that was the highlight.
‍

Christina:

Peak moment. It was great. Also, along that early Replit journey and maybe the middle Replit journey, there were people who, or there was also just a lot of skepticism. How did you think about that when you had smart people being like, oh, you can't do this in a web browser. If you do this, no one will want it. It'll always be slower, nevermind running it on a phone.
‍

Amjad:

Yeah, I mean they also had very reasonable arguments. I remember some of my friends and Jordan and colleagues were like, no one's going to download 10 megabytes of JavaScript to run your thing. Which literally it was 10 megabytes to download the Python interpreter, which was a lot at the time, especially on phones and what have you. I think I just naturally, I’m either wired or through upbringing that it actually just makes me want to prove them wrong if someone tries to knock down an idea that I have. And that's often a good trait. Sometimes it's a bad trait because sometimes they're really right and it's a freaking wall and I can't really get through it. So yeah, I think it just comes naturally to me.
‍

Christina:

How can you tell or how soon can you tell if you're up against a, I dunno, wood wall versus a cement wall?
‍

Amjad:

Over time? I mean this is a question of how do you know ground truth. And it's very hard with things like product market fit, what is the market for this? Because there's all these ways in which you can make assumptions about the market or calculate something or convince yourself of something that's strong. And then I think it's sort of this buzzword first principle thinking or a prior thinking and the idea is what are the basic assumptions about people and markets, whatever that would allow me to arrive at whether this is possible or not, or this is something people want or not. For Replit, the need was obvious. It was solving a real use case, but whether people would pay for it was really hard and painful for a long time. And I think I was sort of steadfast in my belief that I could find a business model or I could find a way for people to pay for it. And at some point, I think over the past couple of years I just realized that when you have alternatives that are free and easy, and when your thing has this mass appeal and it's not made for a specific type of professional and it's replaceable, even if it's not easily replaceable, people will go through the pain. If you're going to charge them, they'll go through the pain to replace it. So most people in the world are time rich and sort of money poor, as it were. So they will always trade off, they'll always put in more time and Replit was saving time. And I just realized that we needed to go up market, go to a place where people actually were time-poor and saving them time is a thing that they would want to pay for. And so Replit Agent is the ultimate expression of that because it actually saves you insane amount of time, orders of magnitude of time, and it could be time that you're spending or time a developer that you outsource to is spending and that could be tens of thousands of hundreds of thousands. And we hear that from our customers all the time. Literally two, three orders of magnitude. And so now it's very easy to capture value. We 10x’ed our growth over the past six months.
‍

Christina:

Congrats.
‍

Amjad:

Thank you. I just try to be rigorous in how I, even before looking at data or anything like that, just try to be rigorous in my thinking.
‍

Christina:

Of what you might want to see in the data or what if it goes this way, you'll think this.
‍

Amjad:

And even I would try to form an opinion before I would actually go and try to research and based again on first principle ideas or what I view, what I know about human nature.
‍

Christina:

That’s neat. You've also talked a lot about hiring high agency individuals. What's that mean?
‍

Amjad:

I think in life, the way we're shaped is to continuously shed autonomy in a way where as a child you're probably the least controlled by nature. You're very open, you're learning things, you're playing and you're breaking all the rules and all of that, but over time you're molded into something that is more acceptable to society. You don't want to get in trouble and then you have all this work and school and all of this work and university and you sort of molded in a way of working that is very, it's the same. Everyone sort of works the same because school applies a certain pressure. I think the high agency people tend to be misfits because otherwise you're going to get this sort of the usual corporate drone. The reason is no fault of their own, but they're not actually—they come up in a system where it sort of punishes original thinking and it rewards conformity. The misfits are those who cannot conform and they tend to be weird because of that. It's not that they don't want conform. A lot of them experience a lot of pain in life, and I experienced that myself. I got expelled 20 times from my school and there was all these, I couldn't sit in class and I kept failing because, my grades were always great, but I kept failing because I just simply couldn't sit in class and we had attendance, you had to do the attendance. And so eventually I sort of hacked into the university to try to change my grades.
‍

Christina:

Did that work?
‍

Amjad:

It worked until I got caught.
‍

Christina:

Are we talking like an hour or are we talking like a semester?
‍

Amjad:

No, a few months I went and bought the gown and was going into all these parties and was just ready to graduate and then I got caught, but I felt like I was owed that, to graduate. I always got A's, but I just didn't want to sit in class. Call it ADHD or whatever, I just physically couldn’t. And so I think a lot of people who tend to be high agency have something where it is harder for them to conform. It's not a hundred percent, but I think that's for the most part. So one thing is you should be open to that. You should be open to people who are awkward. You should be open to people who are dropouts. Obviously now Silicon Valley is actually kind of prefers the dropout, but I think just the openness to being able to interview these people is a big step in that direction. And then having a culture that actually could bring in these people and have them be productive.
‍

Christina:

And so do you find actually the you of today, if you couldn't sit through school, can you sit through meetings?
‍

Amjad:

If I'm interested and if I am running the meeting and I am trying to get something out of it, I'll run it with intensity, but the moment it starts going into circles, whatever, I'll immediately tune out. But I also think that that's bad and you should just leave, right? You should end the meeting.
‍

Christina:

We’ll regroup or when we figure out what we want to get out of this.
‍

Amjad:

Yeah, I'm increasingly doing that. I don't feel like these are productive. When I tune out, I think it's a signal that perhaps we shouldn't be in the meeting.
‍

Christina:

Something is off. What is it about the high agency people or the misfitting? It makes it easy to identify these folks, but why are they effective later because of this?
‍

Amjad:

I think a lot of it is questioning conventions and I think working at a startup. One, if you're trying to build something sort of new and bringing something new to the world, that is actually different than what most people used to do. For example, like everyone codes on their desktop computer, no one wants to code on the cloud or you want to change that. So if they're not wired differently in the first place, the conformist thing to do is actually not to join this company. But that also extends to their day-to-day work where they're more inventive. I think that they're able to look at problems and able to solve it in a way that is novel, whereas maybe the regular engineer would take a lot more time or write a lot more code for it, and perhaps the person who's wired differently would actually be able to look at it at a different angle that is not obvious, and invent something. And we have a culture of invention, we've been able to open source a lot of things that we've done, and we've been able to stay lean for a long time because we have these high agency people.
‍

Christina:

So speaking of doing things new, a lot of things Replit has done over the years, whether it's technical innovations or UI patterns or business models, are new and so good they end up getting copied by giants. How do you feel about that?
‍

Amjad:

I feel really bad, actually I used to feel really bad. I think I take it for granted now that it'll happen. It’s still annoying. It's especially annoying when people do not know that we innovated that. A lot of Microsoft products are designed by Replit.
‍

Christina:

Right. How do you feel being the chief designer of the Copilot suite?
‍

Amjad:

Almost a chief marketer. We updated our mission at some point from making programming more accessible to try to make it more ambitious, to creating the next billion developers. And GitHub is now saying that, so I'm both a chief marketer, chief designer at Microsoft.
‍

Christina:

I’m sure the pay for those jobs is quite good.
‍

Amjad:

I haven't gotten ####, so maybe I should ask for that. But over time you resign you yourself to it. But I think it's good to really understand it because you start to know what is moat and what is something that is easily copyable, what is the lead time before people copy things. You sort of have naturally more urgency because you know that it's going to happen. If it's a good idea, people will copy it. We were the first, kind of, at scale software agent back in September and quickly got copied multiple times.
‍

Christina:

Do you have a model now for what is less easy to copy or what lends itself to more of a durable moat?
‍

Amjad:

UI innovations, you can think of them as public goods.
‍

Christina:

If they're good. They will go off into the universe.
‍

Amjad:

Immediately. Immediately within a week we see competitors copying UI innovations that we do, which is sad because we have a really great design team and we really spend a lot of time and energy stressing every pixel. And then those get copied. Things that are trial through fire or things that have a lot of pain associated with them, especially on the infrastructure side, those tend to be harder to copy. What I said earlier about I love reaching scale because I like designing robust systems. And a lot of times that's what it takes, is that we build a system and we drive as much traffic and growth to it to see it break and fix it and become anti-fragile over time. And I think those tend to have, maybe not a lasting advantage, those tend to have a longer lead time before people catch up to it. It might be in the years. I think Replit infrastructure has many years of, because we've dealt with all the abuse, we've dealt with all the security, we've dealt with everything that has to do with running virtual machines in the cloud and giving it to users. It's a really tough problem and anyone who wants to compete with us has to go through all of that. And then lasting advantages are actually really hard. You know this, anyone who's running a business and then starts to get all this competition, you start to think really, really hard about what lasting advantages are. Obviously the buzzword in Silicon Valley is network effects, everyone wants to build network effects. It is actually quite hard to find a way to build network effects. But I think over time as you're running the company, you start finding opportunities to not only differentiate but also differentiate in a lasting way. And there are not that many ways and you can sort of study them. There's this book Seven Powers, and it just tells you, oh, there's seven ways to build moats. And I think it's very, very true.
‍

Christina:

Any advice for a founder starting out early on this topic? It sounds like you've gotten to a more than before state, but for someone who's early on and it's not yet at that pinnacle of enlightenment.
‍

Amjad:

I mean I sort of tend to be sort of a little fatalistic about how these things go. If I were to go and try to give myself advice when I was 20, my 20-year-old self will not listen, tell me to F off or something like that. And so for me personally, you and I have sort of an experience in that where I was like, okay, maybe I can fund Replit through selling to educators. And you were like, no, it's not a good business. I'm like, I'm going to try it anyways.
‍

Christina:

But it seems like it should be. It seems like it should work.
‍

Amjad:

Well sometimes now it's starting to work with AI. I don't know if people know, but you're the first investor, the first angel investor definitely, in Replit.
‍

Christina:

Coffee shop down here in 2016.
‍

Amjad:

2016. So thank you for that. I think for a lot of people it's hard to learn from experience. However, I think you can teach mental models, frameworks, things like that. It's just this simple thing of what we talked about with seek pain. I think it would've helped me a lot because there are many times a lot of bad decisions that I've done with a company tend to be, I know that this is the path, but it's painful. It requires a pivot. It requires letting someone go. It requires and I try to delay it or try to work around it and almost always, I should have done it like a year or two ago. So if I were to have given myself advice as to kind of lean more into the difficult, and by the way, speaking of sort of moats, and I think if you continuously do the difficult thing, I think you're going to perhaps create something that's quite differentiated because it's the path less traveled.
‍

Christina:

I do actually. I believe that and that, I don’t know about you, but trying to shrink the time where I felt like I delayed making a bad decision. I wish I could tell you. I was like, oh, it's 24 hours all the time. I've got a great SLA. Not true, but at least we talk about it in days or weeks and not months or years. And it's just that it goes down over time.
‍

Amjad:

Yes. Yes, exactly.
‍

Christina:

Well, you all wrote an AI manifesto, why'd you do that?
‍

Amjad:

Well, you want it to put a stake in the ground of Replit’s approach to AI. I think we were trying to hire AI engineers and researchers. We have a great team right now and we have to compete with 5 to 10 times the salary that we could pay. And so we wanted to just tell the world and talk about it internally is that this is what we're doing, this is how it's going to be differentiated.
‍

Christina:

And where did that manifest come from? You, I imagine.
‍

Amjad:

Yeah, me and Head of AI, now President, Michele Catasta. We just spend a lot of time actually, we did this big deal with Google that had an AI partnership and Google Cloud partnership. He was the counterpart, the AI researcher at Google. So we spent a lot of time talking about how AI changes programming, over many dinners and many nights. And so when he joined Replit to run AI, we had already explored what was possible together and we were starting from the same starting points and so much so we had this vision and we wanted to put a stake in the ground. We're also not worried about sharing it. A lot of founders tend to be worried.
‍

Christina:

Yeah, like this is not your moat.
‍

Amjad:

Yeah, exactly. I gave a TED talk in October ‘23 about this is what software agents will look like and this is how Replit will build them. I'm continuously surprise how much we got right. Even the UI back then, we're still kind of implementing, even before test time compute or reasoning models came out. One detail from that talk, I was like, well, when you have a software agent and it's struggling, you can pour more compute into it and it’ll do a better job. And so there's all these predictions that turned out to be right. And I think it's what we talked about with my prediction about where the web's going. It's actually doing the work. Not many people do the work of sitting down and thinking about, okay, where is this headed.
‍

Christina:

And what is your take on where it's all headed?
‍

Amjad:

In AI, there's the idea of the singularity where, it comes from physics, where the point for which you can't predict afterwards. We're not really at the singularity. I don't really believe that. However, I think AI agents are going to take off so much, and they're going to change the economy fundamentally. They're going to change how companies work. One prediction about how companies work is that I think companies, the nature of the firm will change. Companies will be hiring generalists. Because anyone can be a software engineer. You have an AI agent that's quite competent.
‍

Christina:

You can vibe code your way to your translation app.
‍

Amjad:

Right. And you can actually vibe code it your way into becoming a good finance. So you can think about it as in video games, there's this character or hero that can adopt any other hero's superpower. And I think this is what companies will be made from, where high agency, generalist people, that are augmented by AI agents that are able to do a lot and get a lot of work done in any given hour or day or even minutes. I think we're at a point where it's really hard to predict what happens next. I can tell you how the technology will progress and I think that, well, here's a few ideas. In the near future, I think we're all going to get more and more comfortable with software being asynchronous.
‍

Christina:

What do you mean asynchronous?
‍

Amjad:

So right now we sit in front of a computer, we're trying to get a task done and we're clicking and spending time there. We're going to be comfortable with modes like deep research, where you're sending off the computer, the AI, to do a lot of work and come back to you.
‍

Christina:

You go get a coffee by the time you're back and settled.
‍

Amjad:

Or you work on something else. So I think this multitasking sort of ADD way of working will be the way of working in the coming months. Replit, right now, we're making that trade off. We're like, well, we can work for 10, 15, 20 minutes. AI can work for 20 minutes. It could produce 10x the results of working two minutes. There's almost like this linear relationship between how many minutes we're spending on recent tokens or whatever with the output, whether it's quality, UI, all of that. And we're finding, we're still kind of validating that. We're finding that our customers are fine with that. We will send them a notification, they'll put in a prompt, and we'll send them a notification when the AI is done working. In my opinion, that's the main modality of working even by Q3 or Q4, many of us will be kicking off these agent workloads and checking on all them or getting coffee or whatever.
‍

Christina:

Okay. How is this different from long build times though, right? Engineers hate long build times. That's kind of what it sounds like.
‍

Amjad:

Yeah, engineers have experienced this for a long time.
‍

Christina:

But generally weren't positive on that experience. But how is it different in an agentic world?
‍

Amjad:

The compiler taking a minute or 10 minutes or an hour is not going to change the quality of my software. An agent taking a minute, 10 minutes, will fundamentally change the quality of the output. So we're going to be constantly making time-quality tradeoffs. Whereas a compiler, you're just going to have to wait. We actually have a choice. And I think more and more people would prefer, I mean there are always the iterative prototyping modality, but for example, on Perplexity, I'm a big user of Perplexity. I almost have the pro-mode always on, right? And I'll put it in a query, I'll go to a different tab and come back to it in a second.
‍

Christina:

It's so interesting hearing you talk about this because Replit, one of the first things, was like, you get your response instantly.
‍

Amjad:

It’s in the name, loop. Yeah, there's some looping, but you're not doing it. The machine is doing it. Literally all those stuff we built, we just switch out the user to an AI. So the AI is doing fast iteration, but you're not.
‍

Christina:

Yep. That's neat. Another way I think Replit's really different is your approach to marketing. And maybe, I don’t know if you're still combing the web for backlinks these days, but how do you think about it today? Especially, to prompt you a little bit, but community, what marketing people call influencer marketing? How do you think about all of that?
‍

Amjad:

Yeah, so early on in running Replit, I found the power of social media. I was able to bootstrap all these products and businesses and open source projects, like being part of React. Bloomberg never wrote about React. Maybe they did recently, but React became the number one framework of the world. React Native became the number one cross-platform development in the world with very little. Even Facebook didn't market it all that much. So you kind of came into this company realizing power of what going direct.
‍

Christina:

As the kids say today.
‍

Amjad:

Yeah, as the buzzword now. I was like, we used to call it tweeting. Yes, exactly. Now it's like going direct. So that was already obvious to me. But also early on in the company, I hired a PR agency, which they have their uses still, to write about a fundraiser or something like that, they took so much of my time and money and it landed in sort of some outlet that no one maybe, like 30 people read it. And I was like, I look at my Twitter and I'm getting hundreds of thousands of views. And then I literally said that on Twitter that went viral. I was like, oh, I paid this much for a PR agency to do this. I didn't get any readership, but on Twitter it was meta thing about it as that Tweet, it's off, went viral. So that was another learning experience and I think it's more fun that way. And I think just watching the experience of people like established founders like Zuck and others where, and I worked at Facebook where they built him up so much and then they turned him down. And I think that's the business model of media, of corporate journalism is building you up. And then after they finish building you up, there's only one way to go. There's only one way to go to generate the same news headlines. And so if they're not building you up, I mean, I get very little coverage on mainstream media. It's a different relationship that you have. You also just have direct relationship with your users, your customers. I get a lot of feedback, especially on X from customers.
‍

Christina:

Actually before you do that on X or Twitter, you're so much better on that platform than most people, very much me included. But how do you think about it? What are your tips?
‍

Amjad:

The first thing I would say is it's about distilling ideas. So right now you can write arbitrary number of characters. I almost never go above the limit because I think it's a feature, not a bug, to be able to distill ideas. So I might spend, on a Tweet that matters, I might spend 30 minutes trying to distill the idea, and it might sometimes be a background task in my head as I go through my day.
‍

Christina:

As your agent is doing other things.
‍

Amjad:

Yeah, I am trying to distill its most potent potential expression. And the other thing, it's about understanding the mode of the platform. So X is one where you're scrolling really fast, maybe LinkedIn is you're not as fast, you're willing to read. And so with X, you're really having to grab attention in the first few words. There's also a thing about X that, drama is something that is very interesting. So fights are very interesting to people. You don't want to be cheesy and unoriginal about it or overly combative or whatever, but there is a sense in which the beefs work. There's a sense in which also sometimes being a little vague works because people will interpret it in different ways. And then there's a viral effect to that because they will apply their own opinion. Like they'll quote Tweet it with their own opinion or their own view of what you said. Also saying things with confidence, although that might not be how you speak in day to day, I think it's a good strategy for Twitter because that generates strong disagreements and strong agreements, and that actually fuels virality as well.
‍

Christina:

Okay, thank you for those tips. What about community?
‍

Amjad:

Yeah, I mean, I would credit a lot of that to Haya, my co-founder. Early on, she really cared about the people, I cared about the machines more. And so I remember even when she started working on it with me, she was like, well, I want to run a survey to learn this thing or that. I was like, why? I can tell you what it is. I built this thing and she would find out things that I wouldn't have imagined about how people use the product and things like that. And since then, we've just had this approach to our community with this empathy and understanding in how people use the product. I'm only one person tweeting, although I have a large audience. But if you want a lot more people carrying the message of Replit, carrying the culture and vision of Replit, you want to create this core group of users that really care about that, and you want invest in them, and you want to create reasons for them to stay with you, create connections so that people feel like their friends in the community as well. That is somewhat of a moat, that is somewhat of a network effect, and you need to keep watering of the plants. And it's like a constant investment to make, which is hard. And it's also hard to find really great community managers, and the founders need to do that a lot of the time.
‍

Christina:

That's neat. Interesting. You're a powerlifter, yeah?
‍

Amjad:

Yeah. I mean, I used to be more so. I've lost a lot of weight, but yes.
‍

Christina:

How is it similar to running a company?
‍

Amjad:

Well, I think progressive overload is very true. And the idea behind power lifting is you increase the weights progressively and you get stronger, and your perceived sort of exertion might stay the same, but you're able to lift more weights. And I think that is very much true. I mean the amount of workload.
‍

Christina:

Pain you can endure.
‍

Amjad:

Stress you can endure over time. All those things are progressive.
‍

Christina:

You've also read more widely on philosophy than your average founder. Does that influence how you think about AI?
‍

Amjad:

Yeah, I spent a lot of time reading about, just the nature of consciousness and actually trying to struggle with that question. Where I think a lot of people, AI researchers, people in the AI space, are just taking it for granted that the consciousness is an immersion phenomenon. We live in this physicalist reality. Like there's only the physical and so there's nothing to explain about consciousness. It's merely emergent. And I don't really believe that. And I think once you sort of dig into the literature and see how far people, philosophers have explored, they don't have answers. That's the problem. So it's like a slog and then eventually you just still remain with the questions, but at least you've asked the question deeply. And it is a very important question now that we're headed into this world where we're going to have AI based employees and so on, are they actually conscious? If we're going to get to AGI, does that thing need to have rights? If we believe that consciousness is emergent, then what makes us so sure that it doesn't have consciousness? I actually think that my view is it's probably not fully emergent. There's probably something special about humans. And I think is probably a view shared mostly with religious people, perhaps. I am not entirely sure. I can sort of speculate as to what is the nature of consciousness or what have you. But I would say it's more likely than not, that consciousness is not this purely physical immersion phenomenon.
‍

Christina:

If folks are curious to explore some of the ideas behind that themselves, where should they go?
‍

Amjad:

Actually, I said all of that and I am going to suggest a book that actually makes the argument that it is emergent, which is Douglas Hofstadter, I Am a Strange Loop. But the reason it's good, it makes the best argument possible for emergence and it's very digestible and at the end I felt like I wasn't convinced.
‍

Christina:

Okay, so that's good. Neat. We've talked about how your dad encouraged some of your early interest in computers. Alternate universe, none of that happened. I dunno. You never found a computer. What would you be doing?
‍

Amjad:

I was interested in poetry and sort of language, literature. So my mom is sort of the opposite of my dad. My dad is your engineer's engineer, and my mom couldn't be further from that. She is very spiritual. She reads literature and poetry and all of that. She introduced me to a lot of that early on, especially Palestinian literature and poetry, which comes from the sort of struggle and it's very, very potent because it's shaped by that. I don't know if I'll be a poet per se, but I think it's plausible that I would be interested in instead of the crafting of code, the crafting of language.
‍

Christina:

Does some of the Palestinian poetry stand up in English? Is there anything folks should check out?
‍

Amjad:

I would say Mahmoud Darwish is the one that's most translatable to English. He might've even written in his own English, so I recommend that.
‍

Christina:

Great. Well, we're talking here at your office in Foster City, which is not San Francisco, it's Foster City. How's Foster City treating you?
‍

Amjad:

Good I mean, look at the view outside.
‍

Christina:

Beautiful. At least 30 degrees warmer.
‍

Amjad:

Right? It's warm. We have this lagoon and we do one-on-ones while walking around the lagoon. I really like it. The sort of big caveat is there's nothing out there.
‍

Christina:

Is that a feature or a bug though?
‍

Amjad:

It's a feature in my opinion. So there's a Costco people can go to get a $1.50 hot dog there, but we try to have everything here. So in the morning we have someone like a barista making coffee. We have lunch and dinner and some select days, cooked by a chef. And as you know, Haya and I really care about what we eat, care about our health and our employee's health, even sometimes when they don't like it or that we only carry good, healthy snacks and everything, people come up to us like, “Oh, I feel like a lot healthier after I joined Replit.” It is certainly a feature that it is a little boring here. And I think the other thing is we're able to think. I feel like there's just, the space allows you to think a little bigger. I feel like San Francisco, you're just cramped. And then the other thing is, in San Francisco, everyone is, there's such a tight feedback loop between everyone and everything that people say, it's like a bubble, but removing yourself from that has actually netted out with more original ideas, I would say.
‍

Christina:

If a founder came and spent a day with you in Foster City, maybe here, maybe out, but what would you do for a day?
‍

Amjad:

I would do a Q&A with a team. So I’m always trying to bring different perspectives, even if it something that is, someone who runs the company totally differently.
‍

Christina:

Just avoid pain constantly.
‍

Amjad:

Yeah, it's like, the company died. We had Mike Knoop the other day from Zapier and they're fully remote company and they do different things quite differently. We like to hear different perspectives. So introducing to the team, have that sort of discussion, maybe go for a walk, maybe sort of show them some of those very intense meetings we have, show them a few demos that we're working on. I remember one of my favorite visits, that was Jack Dorsey visiting Replit in the SF office.
‍

Christina:

I remember hearing about this.
‍

Amjad:

Yeah, we spent four or five hours together and it was basically all of that.
‍

Christina:

That’s great. Any advice for early stage founders that is this truly unconventional?
‍

Amjad:

There are a lot of rules in Silicon Valley that you can just throw away. When a VC tells you that, this is our process, we write checks only this size, we have to take this percentage of the company. None of this is true. They always break their own rules. So that's one. Two, a lot of what people tell you you're supposed to do—a lot of it is actually not good advice. I mean now it's becoming mainstream, but I stopped doing one-on-ones, for example, like, four or five years ago.
‍

Christina:

So way before it was cool.
‍

Amjad:

Yeah, way before it was cool. What I found is one-on-ones tend to be sort of a therapy session, and I found that while, okay, people need to be able to talk to someone, I just found it to be draining for my energy. People have a lot of anxieties and things like that and I try to talk about them in a group setting, but we also have a coach, so I just send people to a coach too. I'd want to talk about work, I want to understand their worries about work, whatever. But I'm not a therapist and it actually takes away from my time and energy, whatever. So a lot of early stage founders, they find themselves constantly doing therapy and I don't think that's a good use of their time.
‍

Christina:

For those that really enjoy it, excellent. But for those that don't.
‍

Amjad:

Yes, there's all these ideas around who you hire, fundraising, you have to hire an executive team now. Again, you can run companies in so many different ways. I visited xAI think a year ago when they were 70 people, and they told me they had no managers. Everyone reported to Elon, and everyone would send Elon weekly status updates.
‍

Christina:

They probably didn't have one-on-ones.
‍

Amjad:

They didn't have one-on-ones. He would come once a week or once however long, and he would sit and sometimes he would do a marathon where he'd do one-on-ones with everyone or do one to a few people per team. And the cool thing about it is, I know one of the early engineers there who's leading go to market now, so they have purely engineering teams. They don't have non-engineers and the engineers are doing the sales and marketing as well. And so that's a fundamentally different way of running a company. That's one of the most successful founders of our generation, if not the most successful, successful. So I will say every rule can be broken, but try to think through things. It's like, obviously just following every advice is bad. Just breaking everything for no reason is bad, but try to think through things, continue to tinker, continue to change things and that's how you learn.
‍

Christina:

Tinkering in the system is a neat analogy.
‍

Amjad:

And it's hard because people don't like change, but they'll deal with it. It's fine.
‍

Christina:

Okay. Rapid fire around for you. What are you reading right now?
‍

Amjad:

Man, I haven't really read all that much recently.
‍

Christina:

Any good children's books?
‍

Amjad:

That's exactly what I've been reading. Just like a very light read. I read Genius Makers about the recent history of deep learning. I read quite a bit of papers and Hacker News, blogs and internal documents. Erewhon? Have you heard of that?
‍

Christina:

No, I don't. I think I've heard of it, but I've not read it.
‍

Amjad:

It’s by Samuel Butler from the 1800s. And it's very interesting because—are you familiar with The Butlerian Jihad?
‍

Christina:

No.
‍

Amjad:

It's from Dune the movie, there are a group of people that don't like machines against AI and they do this thing called butlerian jihad. And it comes from this book. And in this book there is a tribe or a group of people that find that machines are sort of, dysgenic. It was like at the rise of Darwin. So it's a very Darwinian book. The machines make them worse as humans or less fit, and so therefore they're against any machine. And so it is very interesting for our age, as we increasingly replace aspects of ourself with computing and AI.
‍

Christina:

Yep. This is great, I'll check it out. Do you remember the first program you ever wrote?
‍

Amjad:

It must have been a Logo, a Logo program. Logo is this Turtle you can program. I must've been six years old or something like that. And must be something like move left, move right, go down.
‍

Christina:

What about something where, at the time it seemed like a really big deal but in retrospect it was fine?
‍

Amjad:

After we did the layoff last year, I felt like I couldn't bring back the culture of the company. I dunno if you know this, but there's, there's something about layoff that kind of breaks sort of the bubble or the dream. It was like, oh, we're not always…
‍

Christina:

We're not a family.
‍

Amjad:

We’re not a family, we're not always up and to the right. It's going to be a tough decision. My relationship with the company is not unconditional. I felt like, oh, it's going to be this monumental task to try to bring the culture back and I started doing things that I don't think I've ever done, which is try to do events and things like that we do some of those, but the way we built the culture wasn't that. So I was doing the wrong things. And the way to bring the culture is to win, is to work hard, motivate everyone, crush it. And that's what we did. And the culture is stronger than ever, it’s actually more fun and more exciting than ever. We ended up doing the right thing, but the hopelessness that I felt was like nothing I felt during the company because I cherished our culture so much and I felt like I made a lasting damage to it, which turned out not to be the case.
‍

Christina:

And you didn't need to cosplay something you weren't to get it back.
‍

Amjad:

That's right.
‍

Christina:

What about the flip side of that? What was something that kind of didn't seem like a very big deal at the time, but in retrospect ended up being quite important?
‍

Amjad:

Early on, we were building the Replit platform. We sort of had a choice between making Replit more bespoke, where was, supporting maybe one language or one way of building things versus making a general platform that can run any language or any package and all that. And we went with the latter, perhaps it was more like an aesthetic decision, but fast forward and LLMs are on the scene today. And so what are LLMs trained on? They're trained on standards. They're trained on open source, they know how to use a virtual machine, and all of that. And so the fact that we went with a general, more standard platform way of doing things made it so that you can drop an LLM into Replit and it'll just work. Whereas if you look at some of the, maybe let's look at the low-code, no-code platforms, they're all bespoke. And so LLMs haven't really had a big impact on those businesses as much as it had on ours.
‍

Christina:

Yeah, that's very neat. Well, thank you so much for the time, for the conversation this afternoon.
‍

Amjad:

Thank you for having me.
‍

Christina:

We hope you found this conversation as enlightening and fun as we did. Thank you so much for watching. If you're interested for more with Frameworks for Growth, you can find us across the internet as Vanta and at vanta.com. Thanks so much.