🎙️When Will We Stop Coding?

Hi everyone – hope the weekend's treating you well. Turing Post now has a proper YouTube channel – ‘Inference’, and you can listen to our podcast on all major platforms: YouTube, Spotify or Apple Podcasts.

We’re also working on more video content – both serious and fun. Got something you want us to cover? Just reply to this email and let us know.

What happens when the biggest advocate for coding literacy starts telling people not to learn to code? In this episode, Amjad Masad, CEO and co-founder at Replit, joins me to talk about his controversial shift in thinking – from teaching millions how to code to building agents that do it for you. Are we entering a post-coding world? What even is programming when you're just texting with a machine?

We talk about Replit's evolving vision, how software agents are already powering real businesses, and why the next billion-dollar startups might be solo founders augmented by AI. Amjad also shares what still stands in the way of fully autonomous agents, how AGI fits into his long-term view, and why open source still matters in the age of AI.

Whether you're a developer, founder, or just AI-curious, this conversation will make you rethink what it means to “build software” in 2025.

Watch and listen in this episode.

The transcript (edited for clarity, brevity, and sanity) ⬇️

Ksenia: Hi, Amjad, so nice to meet you. I'm fascinated with your story and the story of Replit (we’ve covered it in our GenAI Unicorn series: Replit: How to Compete Tech Giants by Democratizing AI in Coding). You once championed coding literacy – first at Code Academy as the first employee, then at Replit. And now you're saying people shouldn't even learn to code, right? So what changed?

Amjad: Yeah, that turned out to be a very controversial statement. The main thing that changed was the progress in AI. I've always thought agents were going to be the new wave of programming. You know, code is a means to program – it's not programming itself. And it's actually very tedious. There are all these syntax issues with it and all these problems. And by the way, we've had high levels of abstraction since the start of computing. Programmers used to do punch cards, then assembly, then C, and then JavaScript. I think the next stage is agents.

I used to think it was going to take five, ten years – but with the progress in AI, I think it's going to take maybe two. I did say that some engineers will still need to write code, especially for things that require precision and provability. And if you're going into computer science, I still think it's useful. But for anyone more interested in building applications — you don’t need to learn how to code.

Ksenia: Was there like a particular moment of realization when you understood there would be some people really hard on code and then the majority just doing it?

Amjad: You know, for someone whose identity is almost entirely wrapped in the idea of coding literacy, it’s hard to change that. I think the first time I really felt it was when we were able to build Replit Agent in beta – this was in July or August. We haven’t released it yet, but it was working. And I was like, wow, this could change things.

Then I had this cognitive dissonance. I didn’t want to fully embrace it.

Then came Agent V2 – our version two of the Replit agent, a fully autonomous programming agent. It can work for 15 minutes unsupervised, but we built it in a way that’s very user-friendly and keeps a human in the loop.

After that, I thought: Agent V1 can work two or three minutes on its own. Agent V2 can go for 15 minutes. Now we’re building Agent V3, and we’re hoping to get it to a point where it can work for an hour unsupervised. And if you can do that a few times a year, at some point you’ll have a software agent that can work for a day or two to solve a problem – which is pretty close to how humans work.

Ksenia: What specific technological breakthroughs do you see are still needed to reach that? Are you working on that?

Amjad: Yes, I think there are two things. First, models. And typically with the models, it’s sort of like – I don’t want to say automatic – like computing data. You put in computed data, and you don’t really need architectural or algorithmic innovations for the most part. I think we have what we need to scale the system. We just need more compute, more data.

On the engineering side, you need a platform that supports long-running agents and supports tools and services that allow agents to build applications reliably. You need to be able to test things as they go. You need to be very good at prompting. And I think there are ways you can make the agent more reliable by providing a platform that has certain features – for example, being able to fork and try multiple things at the same time, like doing sampling or test-type computing.

So on the platform side, it’s similar to how you build a platform for humans, but not entirely. There are a lot of challenges with building a platform where a model can be embodied to become an agent.

Ksenia: Can you talk a bit about these challenges? Because, you know, the year 2025 was named the year of agents, but still agents do not really work that well.

Amjad: Have you tried a Replit agent?

Ksenia: No.

Amjad: It works well. Yeah, people are running businesses on it. Like entire businesses, people are making money every day.

Ksenia: What is your best case that you're really excited about on Replit?

Amjad: A large real estate business used Replit to build an algorithm for routing leads to real estate agents that created a 10% improvement in conversion, which is millions of dollars. That's on the bigger side. On the smaller side—I don't know, like every day I see one. The one on my mind right now: an Airbnb property manager who runs his entire back office and backend in Replit to run his business. So it spans anywhere from small businesses to big enterprises.

Ksenia: Do you reposition Replit anyhow, like from coding platform to something fundamentally different with, like, a no-code idea?

Amjad: You know, the problem is people call it vibe coding, and that sort of stuck –which is not the best term for us, because you don’t even look at the code. But I think the term is useful, so we’ll probably still have to use it for a while. But again, in the next year or two, it’s just going to be so abstracted away from code that...

The problem is, no-code has certain connotations that aren’t right either. No-code tools have always had these limitations that are very annoying. Like, if you’ve used them, they let you build something quickly, but then it’s really hard to iterate. That’s why having an AI generate code is more flexible – it’s less lock-in.

So I don’t know what to call it. I mean, the best term is probably software development agent.

Ksenia: Software development agent… Well, I don't like vibe coding either. I'm not a coder at all, but you know, when I'm trying all the different AI tools for that, it's like I'm just texting with a machine and then it builds it. I don't know the term – maybe you can come up with a good one.

So let's go back to AI agents and the companies building businesses with Replit. You once said that AI agents will create billionaires, right? And what we're talking about with AI, with AGI, is that a lot of wealth is supposedly going to be created.

What are we doing in a world where coding is no longer a barrier? How will these companies be created – will it be easier, harder, or just completely different to build a massive business – from an entrepreneurial side?

Amjad: I would say we're already seeing results. It's not exactly agents, but we're already seeing AI create highly efficient teams, generating multi-million dollar revenue per employee. Even Replit — the revenue per employee is many multiples better than a traditional SaaS company.

Like, the best, 99th percentile SaaS company is nowhere near the best AI companies in terms of revenue per employee. I don’t want to talk about ourselves, so I’ll give an example that’s not us. If you look at Cursor, for example, they achieved a $10 billion valuation with, I think, 50 people.

That’s really unprecedented. So that gives you a hint of where we’re headed. And I think those efficiencies will keep happening – until we hit the point where agents are truly autonomous and can work for days, weeks, or even months uninterrupted. Then you’ll see entrepreneurs deciding not to hire people. Or hire just a few. Keep it like a bootstrapped small business and still be able to generate an insane amount of wealth. A lot of revenue for themselves, maybe.

I think a billion-dollar valuation for a 10-person team – something that looks more like a small business than a venture-backed startup – is absolutely possible. Even in the next year.

Ksenia: I think there are two different things that we're talking about. There are GenAI startups like Cursor. And Replit is a little different in that sense because you've been around for almost a decade, right? So you've been gradually growing to unicorn status before GenAI went crazy.

Amjad: But I think every startup has to be a GenAI startup. Back in 2010 or 2009, we used to say "a cloud startup" or something like that. But now, every startup is a cloud startup. No one is actually racking servers or renting VPSs. Everyone's using virtual machines and containers on the cloud.

Same with "mobile-first" startups – you don't hear that anymore because everyone has a mobile app. And I think it’ll be the same with AI. AI will be in every company, and the idea of an "AI company" will just cease to exist.

Ksenia: But you create tools for people who can now build businesses with AI. So, if you can extrapolate from what you see – how people use Replit – what will the businesses of a one-person or five-person company look like, if they can build a billion-dollar company with AI?

Amjad: So there's an older story before Replit agent, but there was a teacher who saw a problem he wanted to solve and built a bunch of tools for teachers to generate lessons, do assessments, all that. Got to hundreds of millions of dollars in valuation in a year or two.

Of course, he grew the team – it wasn’t like a solo founder. But I could see the future version of him being more of a solo founder or a bootstrapper with five to ten people.

Right now in Silicon Valley, a lot of founders are just trying to predict, trying to research and figure out what kind of problems they can work on. But in a world where everyone has the tools, then everyone is empowered to solve their own problem – or their community's problem. And that creates a lot of value in the world.

Ksenia: Let's talk a little about AGI, since you expressed interest in philosophy, the philosophy of mind, and AGI. What's your personal take on AGI? What exactly do you think it is, and how close are we to achieving it?

Amjad: It's always hard to discuss these things because ultimately there are unknowable elements – like the nature of consciousness. But my intuition and belief is that it’s perhaps harder than AI researchers, labs, and the common consensus in Silicon Valley assume. The belief that AGI is going to be human-level or human-like feels optimistic.

The true definition of AGI, the one the original AI researchers talked about, is being able to drop an agent into any environment and have it successfully plan and learn skills efficiently on the fly. That’s how humans work. Obviously, we have varying degrees of ability, but if you drop me into a basketball game and I know nothing about basketball, I might observe people for a bit and then deduce how to play. Or someone could teach me for an hour, and I’d pick up the basics. I won’t be great, but I can start learning.

Whereas with AIs right now, if you drop them into a totally new environment, they tend to not be able to do anything.

Because of how they’re trained, we haven’t seen AIs that truly generalize beyond their training. Now, modern startups and labs try to change the definition. They’ll say, “We’re building a remote worker that’s human-level in certain domains.” Sure, that’s going to happen – I'm already seeing it happen in coding.

And yes, if you get enough data for most types of knowledge work, you’ll be able to create a general-purpose remote worker. But is that AGI? I don't think so. Because it’s not generalizing or learning efficiently. And those are the breakthroughs we need to actually reach AGI.

Ksenia: I haven't noticed any doomism in your interviews or anything like this about AGI.

Amjad: We need to understand the nature of the mind. We need to understand consciousness. We need to understand the brain a lot better to be able to get the final breakthrough, I think, in simulating a mind in machines.

No, I think there's a big risk in AI like in any other technology, but I don't think there's a Skynet-type risk – like in that movie Terminator, where a single AI entity takes over the world. I don't think that's plausible.

Ksenia: How does thinking about AGI fit into your broader vision about Replit and its development?

Amjad: Broadly speaking, we want to create the best habitat for AI. We want to create the best set of tools, of computing primitives for AI. And so we want to be able to drop the state-of-the-art model into Replit and have it perform the best, and have the best user experience on top of that.

If we get to AGI, I'd hope that it would perform best inside the Replit environment. That being said, it might become so good – if you achieve AGI, it becomes so good, so intelligent – that it can construct its own tools and wouldn’t need us.

And so part of the reason why I don’t like to sit down and try to predict AGI is because there are certain outcomes we don’t have control over. If we actually achieve AGI and then ASI, which is superintelligence, then it doesn’t matter what we predict, because the world is going to rapidly change. We’re going to reach the point that’s called the singularity.

So I’d rather prepare for what I think is the more likely: AI that is continuously improving and becoming more autonomous, but is generally still a tool.

Ksenia: So mostly people augmented with AI. You are famous for open-sourcing stuff. What role do community and open source play at Replit?

Amjad: There are a few ways to build programming environments and agents. One way is to say, we're going to build internal systems that can be automated. And that’s what a lot of the no-code systems are doing. Whereas with Replit, we're like, okay, we're just gonna give the agent open-source tools and open-source languages and have it create apps based on the open-source stack. I think that's better for the agent. I think that's better for the world because the applications are standard applications that run on basically any Linux machine.

As for community – community has been an integral part of Replit’s growth. People shared applications, remixed them, did a lot of really great things in terms of collaboration. We have a way to list bounties or problems on the site and allow other people in the community to help you with your code.

But I don't think we've achieved the true potential there. And I’ve always imagined a world where there's an ability to transact naturally inside a platform like ours – where agents can provision services and pay people in the community based on those services. I think there's a much better way to compose a software community –
one where there's a win-win-win scenario: win for Replit, win for the community, and win for the end user.

Ksenia: What's your take on MCP and A2A? Is it going to be a protocol war or are they going to complement each other?

Amjad: I think MCP won already. With these things, you only need a slight edge with an open-source protocol and you're just zooming. It's really hard to catch up. Like Bitcoin and Ethereum. Ethereum will never catch up, even though you could argue it's a better system.

There are AI systems now that claim to be better than the Transformer, but it's really hard to catch up because the Transformer is good enough and everyone's investing in it. You just need a slight edge and you're off to the races. There's a first-mover advantage that accumulates over time. So I think MCP won.

Ksenia: My last question will be different from what we've discussed before. You are an avid reader. What is a particular book or idea that profoundly shaped your vision of how you think about AI and AGI?

Amjad: You know, a lot of people in tech read Douglas Hofstadter. There's a book called Gödel, Escher, Bach, which is a very interesting book about AI and consciousness and mind, but it's very winding and requires a lot of work to go through. But there's another book called I Am a Strange Loop. And his whole thesis is that consciousness is an emergent component of a system that can self-reflect.

So the moment you can loop on yourself and self-reflect and have a model for yourself, you're actually conscious. And he makes a very good argument for it. He uses a lot of analogies to arrive at these conclusions.

But I actually ended up being less convinced. I think it's a great book, and it’s one of the best attempts to explain consciousness within the current scientific consensus. But I think you can come out of it either fully convinced – and that would make you someone who thinks AGI is going to come very soon, because it seems like you just need an AI that knows how to reflect – but yeah, I ended up being less convinced, that’s all. But it’s an amazing read. It’s very poetic, too.

Ksenia: The last question. What excites you most about this AI-powered world you're helping build?

Amjad: I think access to opportunity. One of the paradoxes of the internet is, it was supposed to be this great equalizer. No matter where you are or what resources you have, you should be able to contribute to the internet and make money on it. But instead, the wealth really concentrated in Silicon Valley and this one geographic area.

I think part of the reason is, the skills and capital needed to build anything are still very high. And with AI, that’s going to collapse.

It’s going to be deflationary on many things, including creating software. So the costs will go down. And that means there’s going to be more opportunity for people in emerging markets to be tech entrepreneurs and build great businesses.

Ksenia: Yes, I certainly hope so. Thank you so much. Thank you for your time and thank you for joining me today. And to our readers – thank you for watching and reading!