When do you think a solo entrepreneur will be able to build a billion-dollar business in the next few years?
This is Amjad, founder of Replit, a platform that turns your ideas into an app. Our mission is to make not just software, but entrepreneurship itself accessible, because building your own business is one of the best feelings in life.
But let's be honest, is it really possible to just sit down and build a billion-dollar company with AI? Amjad says it comes down to three simple steps. They used them to grow Replit to 3 billion. I tried these steps myself, and it didn't quite turn out as I expected.
Hello everyone and welcome to my channel. Today I have an amazing guest. Amjad, founder of Replit. I've talked about Replit on this channel a few times already.
Because I use it myself. I'm amazed by your story. I want to start with a question. You said you want to open up new opportunities for a billion programmers and developers in the next couple of years. But at the same time, I heard you say that in a couple of years, companies won't need developers. Can you explain that?
Yes, I'm mostly talking about entrepreneurs like us. Large companies will always need software developers, but people who have an idea—and everyone has an idea, you can just go out on the street and ask anyone if they have a business idea. Everyone has an idea, but most are held back because they don't have technical skills or someone who can do it for them. As a programmer first and foremost, my friends constantly asked me, 'Hey, can you program this idea for me?' Well, now they can do it themselves. We're approaching a moment when anyone can start a business. It's hard, the technology still needs to mature. But we already have a ton of stories of people building the app of their dreams that they've been harboring in their heads for 20 years. For example, one financial director at a venture capital fund. He's a true expert in his field. He knows very well how to manage a venture fund. But he couldn't find the right tools for his work. He had ideas on how to build them. And you know, finding engineering resources is almost always difficult. So, he built his dream app on Replit in 3 months, sold it, and got a bunch of contracts. I think he's already on his way to 5 million in annual revenue. He quit his job, now he's an entrepreneur. And he keeps saying, 'Well, at some point I need to hire a software developer. Maybe I do.' But he still reached 5 million dollars in revenue without doing that.
Yes. I'm sure that at Meta, OpenAI, and at Replit, engineers will always be needed. But there could be much more entrepreneurship, many more businesses in the world if we remove this bottleneck of software development.
You have a screen here that displays the main metric. What is the most important metric right now?
Yes, every team has its own screen. Some have infrastructure metrics, some have product metrics. It all depends on the team. But almost every team has its own screen.
So there isn't one common metric that everyone looks at?
Well, annual recurring revenue, everyone is responsible for that.
Can you share the current number?
160. Amazing. You're growing just insanely. Your growth chart is that hockey stick. The dream of Y Combinator.
Yes, especially if you build it for 10 years.
Well, yes, that's how it will look. It's so motivating for entrepreneurs who don't see progress right away.
How many users do you have now?
Let's see now. About 50,000.
And do you know how many apps built on Replit are actually live right now?
Yes. About 350,000 paid apps online. And it's growing fast. 25% a month.
And do you know how many of them are actually generating revenue?
No. We only have individual stories that reach us for now, but pretty soon we'll allow Stripe integration to monetize, and then we'll be able to track that.
I think I already connected my Stripe.
Well, yes, and the agent can already do that, but we actually want to make it even simpler, in one click. Our mission is to make not just software, but entrepreneurship itself accessible, because that's what changes lives the most. You can build some program and yes, it's cool. But creating a business is, I think, one of the best feelings in life.
By the way, I'm trying to build something in Replit right now. And here's what I ran into. Yes, it builds a beautiful interface, but then... Ah, look, the service is unavailable. Here's a bug right now. It feels like... Ah, interesting. It's on the deploy. You can go here into the logs and see why the service is unavailable. See, there's an error. Copy it, give it to the agent and say, 'When deploying, I get this error.' But in general, I understand that you still have to work.
Do you have data on how long it takes on average to build something that actually works? I'm trying to make a tool that will analyze my videos, and anyone will be able to analyze their YouTube videos and find those that have potential if you change the title and thumbnail. You know how it is. You repackage a video, and they suddenly start getting a ton of new views. So I'm trying to automate this, but I've already spent 6 hours on it. It's a whole process.
Yes, it's a process. You're still acting like a developer, or rather, like a development manager. You have a very capable but easily distracted intern. And you need to manage them very skillfully. For example, you write a prompt, it's one sentence, but I would spend another minute or two and write it like this: 'When I deploy the site, I get this and that error, but in the preview everything still works.' The exact wording is very important. You know, writing prompts is basically not that different from programming. We just removed the syntax from it. And you don't need to fully understand the syntax and a bunch of details under the hood, but accuracy is still needed. By the way, this also helps in communicating with live developers. An app like yours, for example, would take, I don't know, a couple of days, but before, even a senior-level engineer would have spent a couple of weeks and 10,000, 15,000 dollars on it. That's about what it would have cost you, but like this, 2-3 days, I think you'll build it.
How do you learn to make more accurate prompts?
We have a YouTube channel. We have a great developer advocate. He creates a lot of content. His name is Matt, and we teach people prompting and how everything works inside. Education is in Replit's DNA first and foremost. When we talk about a billion developers, that billion needs to learn. It won't come on its own. And learning is non-linear. And you need to be very resourceful. Go to YouTube and search for how to prompt. You spend a lot of time practicing, building something, changing the style, for example. Someone goes to ChatGPT and takes, for example, the O3 model or the thinking GPT-5, gives it an idea and says, 'Hey, turn this into a really cool prompt for me.' You know, Paul Graham, founder of Y Combinator, the best accelerator in the world, wrote an essay about being resourceful. He talks about the quality of founders. And one of the most important is being relentlessly resourceful. Founders who possess this quality can find resources to remove all obstacles in their path and not hit a wall. Because entrepreneurship is largely about hitting these walls and breaking through them. And here, the analogy with a video game helps. In games, especially open-world ones, you constantly run into tasks, and it's unclear how to get to the next level. It's creative thinking. Wander around, find clues. Entrepreneurship and software development are structured much the same way.
Tell me, what changed when you decided to start a company? How did the thought come to you that I need to build this? Here you need to separate launching a project and launching a company. Launching a project for me was obvious. Programming is hard, so it needs to be simplified. That's very technical logic. But launching a business was not so obvious. I worked in startups and knew how painful it is. I worked, for example, at Codecademy when I first came to the States. It was built on the open-source version of Replit. And I saw how hard it is, really a lot of work. So Replit remained a side project on the side and grew in parallel. And I didn't want to turn it into a business. I even wanted to sell it to Facebook when I worked there. And I wanted to stay there. I was comfortable, I was happy. And to leave a job and dive headfirst into your own thing is very scary. But the risk was mitigated because we already had a platform that people really loved. 100,000 users a month and more. And after a lot of deliberation, after you try to understand what's important to you, what fills your life with meaning—serving clients, fulfilling a mission, helping people build businesses—it all felt very, very important. So it's largely about meaning.
So with a tool like this in entrepreneurship, there are two problems left. Before there were three: come up with an idea, build the product, and sell. Now it's just come up with an idea and sell.
Exactly. What do you think will happen when everyone starts building apps?
Yes, when everyone builds an app, it's still a skill, right? It's a skill you can develop. And here, persistence, resourcefulness, persistence is very important—not giving up after 6 hours, spending at least another day or two. And also, I think expertise in your subject is very important. For example, you have strong expertise in YouTube, and you need to upload that expertise into the agent. That is, you will prompt in a certain way, as if you are uploading your knowledge. That is your competitive advantage.
But at the same time, OpenAI models are trained on what I know. And yet, they sometimes already determine better than I do how to make a good video on YouTube.
But you still have hidden knowledge that isn't explicitly expressed anywhere, not in your videos, nor in any content online. That financial director from the venture fund, he has knowledge and skills that he accumulated over years, and he can turn them into an app. And you won't find that in any blog, in any article. Each of us, living life, accumulates experience that LLMs simply don't have. Just because they aren't materialized.
But do you think there will be a time when artificial intelligence itself sees a problem, comes up with a solution, writes an app, and a person won't be needed at all?
You know, I probably stand out a bit from the general Silicon Valley context. I'm quite skeptical about the idea of AGI. I think we can build very competent agents, but a person will still be needed as a driver. Because, as you understand, LLMs are structured in such a way. They are trained on the entire corpus of texts on the internet, in books, just everywhere. And these are texts of the past, of what has already happened, of what people have already written. But can they come up with new ideas, creative ideas? Ideas based on what is changing in the world right now? They don't learn continuously. They are in a closed box of what they were taught. It's almost like a library in a very deductive sense. This library has a bunch of ideas. It can mix them, combine them, but a truly new idea. Something that no one has seen before, that, I think, is still a territory where humans have a special place.
I really like your answer, but my inner fear wants to object. Yes, yes, please.
All ideas already exist in the world. When they make a new movie or write a new book, they take an existing idea and just change the characters, change the circumstances. A new idea is always there. Take Bitcoin, for example. You're into crypto, Bitcoin.
So Bitcoin was built on a twenty-year history of attempts to create digital money. And in its bibliography, it references Hashcash, the Proof of Work algorithm. And this Proof of Work already existed. The idea that a machine solves cryptographic problems to protect the network. It was originally invented to fight spam, when there were no good filters yet. And the idea was: 'I send an email, solve an expensive cryptographic problem, just to show you that I'm not a spam bot.' And Satoshi Nakamoto took all these existing ideas, you're right. And put them in a new package, but he added one new idea on how to solve the double-spending problem. And that's the blockchain. You know, it's easy to talk like 'nothing new.' But if you look closely, really, look at what Einstein did with his theories. There is always one new insight, one very powerful new insight. I would even say it's almost like some kind of divine intervention. In the appearance of truly new ideas, there is something spiritual. I'm sure you've experienced that too. And I think it's fundamentally human.
Yes, and also that feeling that it's you who must bring this idea into the world. I think artificial intelligence lacks that. Okay, let's see what's happening here. Let's go to the overview and try again. No.
Okay. Click here. It's asking to redeploy. Let's wait a second while it deploys. It's writing that it fixed the problem, by the way. Okay, while it's deploying, you said that the gap between a Replit user and a senior Google developer will disappear in 2 years? Is it still worth learning to code? And what will happen to the developer profession?
I don't think this profession will disappear, especially in very specific fields. It's about what we were talking about. There's a lot that is poorly represented in data. If you're, for example, a platform engineer at Google and you're maintaining a service for a billion users, you have knowledge and understanding that you got on the job and that LLMs simply don't have, because no one has written it down anywhere. That's that hidden knowledge. And such engineers will always exist. If you're an engineer at NASA building fault-tolerant systems or verified systems. For example, I don't want my Tesla's autopilot to be made with vibe coding. You understand, there are many systems where lives are at stake. And for that, we want to have engineers who work at a very grounded level, practically mathematically precise. So, situations where engineers are still needed, there are many of them. But if you're creating a product, just go and build it, don't wait. If you need to learn a little code along the way, well, learn a little, but your goal is to build a product. I would start with building. And as we've already discussed, resourcefulness helps a lot along the way.
And what should developers who build these kinds of apps do now?
Oh, it's going. Okay, we're logging in.
And again... now Google authentication. But this is how developers work. This is what the iteration looks like.
So you solve one problem, and then the next problem arises. And will it ever get to the stage where I won't have to do this?
Yes, because it could run and test everything itself.
Yes, we... that's what we're working on.
Yes. Can you talk about it or not yet?
Well, actually, not yet, but I'll give you a couple of hints.
Any coding platform today automates code generation, and that's great. But all the routine, non-creative, and annoying work remains with you.
Exactly, control and testing.
It's quality control and testing. That's what we're solving.
Okay. I create a lot of content in English, but I actually duplicate a lot of it in Russian. I have an incredible newsletter, I think, where I talk about all the tools I use. I go to these conversations about once a week. I have some cool guest, and they tell me something, like, 'Marina, try Poppy Computer. Marina, try Claude. Marina, try this.' I come home, try everything, and some things stick, some don't. About what sticks, I write in great detail. If you want to find out, for example, what 30 things I've automated with artificial intelligence, there's a whole database, like school emails, tracking analysis of my high cholesterol. I've described all of this in one letter. You can subscribe, get this letter, and we'll be in touch constantly. And not depend on algorithms. So, are you launching something soon? Is there pressure on the team? How are they handling it?
Yes, we have an offsite planned for next week. We're going to Los Angeles to the beach. At least the atmosphere is relaxing, but we're diving deep and just working hard. A week later, we'll be back here. It's called a sprint week. We do it before every launch. People usually work 14 hours a day.
