Replit’s $9B perfect storm, and “AI psychosis”
A couple of years ago, Replit was flatlining at a few million dollars in annual revenue and running a mostly empty office after a layoff. Today the vibe-coding pioneer is valued at $9 billion, up from $3 billion just six months ago. Recorded live at the Masters of Scale Summit in San Francisco, CEO Amjad Masad joins Bob Safian to ride out the AI storm, revealing what made the turnaround possible, including a slow and steady enterprise bet while competitors poured money into ads, and what it really means to be an AI-first company. He weighs in on what he calls AI psychosis, and says cybersecurity, not AI extinction, is the risk worth focusing on. He also argues that automated software engineering is already safer than human engineering, and describes how that’s changing work inside Replit, where the clatter of mechanical keyboards has given way to whiteboard debates.
Table of Contents:
Transcript:
Replit’s $9B perfect storm, and “AI psychosis”
Note: Transcripts are automatically generated from episode audio, and are not fully corrected for spelling, grammar, and formatting.
AMJAD MASAD: A year ago, if I’m walking around our office, you hear just keyboard clicking from engineers. They like these kind of mechanical keyboards.
BOB SAFIAN: Yes.
MASAD: They’re really annoying. But now I walk around the office, and it is actually very dynamic. It feels like a New York creative agency, almost. Everyone’s on the whiteboard. People are arguing. People are talking. I don’t think our place in the world is to sit down and act like machines. The second-order effect of making software more accessible is that we change the future of work. Work becomes a little bit more human. It becomes a little bit more creative, and we leave the mundane tasks to the machines.
SAFIAN: That’s Amjad Masad, CEO of vibe-coding pioneer Replit. Replit has seen an extraordinary boom, from a near-dying business in 2024 to a billion dollars in annual revenue today. In this episode, recorded live at the Masters of Scale Summit in San Francisco, Amjad shares what made Replit’s turnaround possible, including what it really means to be an AI-first company. We also talk about what he calls AI psychosis, how agentic engineering will change the future of work, and more. So let’s get to it. I’m Bob Safian, and this is Rapid Response.
[THEME MUSIC]
There are a cohort of AI companies that have seen valuations double, triple, and quadruple. Our next guest is running one of those companies. It’s the vibe-coding company Replit, which went from a $3 billion valuation to a $9 billion valuation in six months. It’s on track for a billion dollars in annual revenue. In this session, we go into the eye of this part of the AI storm with a leader who’s both riding the maelstrom and navigating it. Please welcome the CEO of Replit, Amjad Masad.
MASAD: Thank you.
SAFIAN: Come on in. Come on in. I wasn’t sure if you’d be wearing your rain slicker after my AI maelstrom analogy. I’m going to use the idea of an AI storm playfully to help frame our conversation via storm movies.
MASAD: All right.
SAFIAN: Their titles reflect the journey of AI companies, but without the disaster element,
MASAD: Like thunderstorm.
SAFIAN: because I know that Replit is thriving. So are you ready?
MASAD: Let’s go.
Copy LinkFrom long conviction to breakout moment
SAFIAN: We’re going to jump right in. All right, first up: The Perfect Storm. The growth you’ve seen has been extraordinary. It is like a perfect storm, everything sort of coming together in extraordinary and unexpected ways. I know you put a lot of work into it. From the inside, what does that feel like? Is it exhilarating when everything is growing so fast? Is it terrifying? Do you have to pinch yourself to say, “Is this real?”
MASAD: It’s all of the above. The first feeling I had was a little bit of vindication,
SAFIAN: Uh-huh.
MASAD: because I started talking about everyone being able to code and create software back in 2010, when I was building the kind of precursor to Replit as an open-source project. Then I was a founding engineer at Codecademy, where we taught a lot of people to code. And when we started Replit in 2016, we were like, “We want to blur this distinction between learning to code and doing code, and build an environment that really makes it so that anyone with an idea can build software.” And for a long time, people didn’t believe us. Raising capital was hard. Recruiting people was hard.
SAFIAN: Because some of the technology wasn’t ready yet,
MASAD: It wasn’t ready,
SAFIAN: the way it is now, right?
MASAD: Yeah. It was on track, and I could see it happening. There were many moments where the company could have died, or I could have quit. It would have been very easy to quit at many different moments. And so in 2024, when we invented what would later be called vibe coding, when we came up with the first fully natural-language-based programming environment, there was nothing out there at the time. People really appreciated it, and we started seeing these stories of people having ideas. One of the early experiences I had was someone recording a video of themselves and posting it on Twitter and YouTube where they were crying because they had this idea of overlaying all their memories and their photos on a map and being able to explore their life through this map. It’s a simple idea,
SAFIAN: Yeah.
MASAD: but it’s a personal app. It’s not even a commercial app. And he was crying because he said he had this idea for so long. He tried to do it with all these different technologies, and for the first time, he was able to do it. So it was exhilarating in that sense, giving people superpowers. And yeah, there’s a sense of, is this lasting? How do you actually make it defensible? How do you make a real business out of it? Will it go away? Before that, I’ve also been an angel investor for a while, and I’ve seen some early AI companies do the hypergrowth and the hyper-deceleration within a year. So there’s a fair amount of paranoia built into how we operate. In many ways, you hit product-market fit. It should be relieving, but in that case it wasn’t. It was kind of the opposite.
SAFIAN: Yeah. When things are going that fast, which they are for you and your business, right?
MASAD: Yeah.
SAFIAN: Because they’re growing that fast. I know you grew up in Jordan.
MASAD: Yeah.
SAFIAN: You emigrated to the U.S. in 2012, and as you say, you spent years grinding away as you were —
MASAD: Yeah.
Copy LinkHow dark moments shaped resilient leadership
SAFIAN: Trying to find your groove. In those slower times, are there lessons from those times that you go back to now as a leader in the midst of the noise you’re in now?
MASAD: Yeah. There were many moments where it felt darkest just before we were able to invent something big or new. Even just before we invented agentic coding in 2024, we had to do a layoff. We were six or seven years into the company, we had two or three million in ARR, and it was flatlining. It wasn’t growing at all.
SAFIAN: Two or three million?
MASAD: Yeah.
SAFIAN: Two or three million, just a couple of years ago. Now you’re at a billion.
MASAD: Yeah.
SAFIAN: I want to pause and acknowledge that. That’s crazy, man.
MASAD: It is wild. And we were burning a ton of money. We hired a great team, but the future hadn’t arrived as fast as I thought it would.
SAFIAN: Yeah.
MASAD: So I kind of mistimed it a little bit. Unfortunately, we had to do a layoff. Around the same time, we also upgraded the office. We went from a small office in SF to this really massive office in Foster City because the space was so cheap there. We had to do a layoff at the same time, and it was this big office, but it felt dark and cold because it wasn’t really occupied. Every day, I would go into the office and sit at my desk and wait for another person to quit. The thing about doing a layoff is you also burst a bubble. You burst a dream.
SAFIAN: Yeah.
MASAD: I’m sure a lot of people are familiar with Steve Jobs’ reality distortion field. Really, every founder needs to create some reality distortion. You’re selling a dream, and suddenly you’re saying, “We’re running out of money.” The dream immediately dissipates.
SAFIAN: I’ll bet those people who quit are now saying, “Oh, can I come back?”
MASAD: Yeah, some of them came back, actually, and we’re always welcoming people to come back. But there was this juxtaposition where, in this cold, dark office where everyone quit, we had a war room. Whenever we have a big idea we want to pursue, we take this YC Demo Day mindset of putting five, 10, 12 people in a room, working really hard toward a milestone, and doing it in a few months. There we were inventing agents. So I would go into this room, and there was enormous energy and excitement. People were seeing the future on their screens for the first time. We saw this entity coming to life that was able to code and run the code and deploy it. Then I would go back out, and the rest of the team felt total despair. I think the lesson from that is there’s always hope. There’s always something you can do to turn things around, and there’s a group of people at your company who are going to be motivated to do that.
Copy LinkWhy enterprise trust became Replit’s edge
SAFIAN: Let’s go to the next movie we picked here, which is Twister. From the outside, it can seem like, from the time you were at a couple million dollars to a billion, things just fell into place, like everything just sort of came together. But there are inevitable twists on that road, choices you’ve made. As you’ve navigated that upswing and tried to stay in the sweet spot, what’s been most important? What’s most important now in keeping that moving?
MASAD: Yeah. After the whole vibe coding hype took off, a lot of people entered the space. Maybe they brought different approaches to vibe coding; maybe they were just copycats. That happened too. It was annoying and difficult, but what always felt important for us was the businesses being built on Replit and operated on Replit. There was something that felt defensible about the way we architected Replit to be reliable, scalable, and secure. That felt really important. At the time, no one cared about security, so we’d go out and write about how we’d built things to be more secure, how vibe coding had to be secure. Otherwise, we’re going to get into a world where all the software we’re building is really vulnerable, and we’re seeing that now with a lot of vulnerabilities coming online. Not many people cared except the people who were seriously doing commercial activity on Replit, running a business, or, in an enterprise setting, adopting Replit. So we decided to focus on those people. Many of our competitors were full-on marketing, spending perhaps 20 or 100 million dollars a month running ads and doing all of that. But we decided to take the slow-and-steady approach of going to our customers and saying, “OK, what can we build for you? How can you, American Express, for example, be more comfortable rolling out Replit and building mission-critical applications?” That was a very unsexy type of enterprise work.
SAFIAN: And yet it wasn’t slow and steady. You describe it as slow and steady, but the impact wasn’t.
MASAD: Yeah, the impact wasn’t, because that started compounding as well. Our enterprise business started compounding, and now it’s the primary focus of the company: small and medium-sized businesses and enterprise businesses, people who are serious about making reliable software, scaling it, and relying on it to run their business.
SAFIAN: I know you’re a believer that the idea of vibe coding is great, that everyone could be able to code.
MASAD: Yeah.
SAFIAN: And you can code in natural language. I’m curious: among your own team, you’ve got engineers who are building a product that essentially can make their job obsolete in some ways.
MASAD: Yeah.
SAFIAN: Do you have conversations about that?
MASAD: Yeah. There’s somewhat of an inherent tension inside the company, for sure. Some of our engineers are motivated by that. You’d be surprised. We have a few AI engineers who joined the team saying, “I want to automate my job.”
It’s like, “What kind of motivation is that? Do you hate coding? Is that why?” But everyone inside is really excited by the impact we’re having on people’s lives, like the entrepreneurs we’re seeing build businesses that otherwise wouldn’t have been built because they suddenly have that power. That’s really what gets us excited. I do think we all face this question. Every knowledge worker essentially faces this question of, “The things that I’m doing might lead to my replacement,” because you’re generating data that is getting trained in the AI models, and that tension exists in tech broadly.
Copy LinkWhy AI security matters more than hype
SAFIAN: Yeah. In the tech neighborhood, you’ve decried what you call AI psychosis.
You’ve called out Dario Amodei specifically at Anthropic for overheated predictions about AI-fed job losses.
Today’s fear over AI agents run amok, is that also psychosis? Or is that a different AI risk?
MASAD: No. In fact, that is really the thing I’ve been talking about, the risk I’ve been talking about. I’ve been on the cybersecurity risk for two years now. I think everything else is a distraction from that.
MASAD: When they go out and talk about how AI is conscious, and they go meet the Pope and do all this random stuff, it’s really distracting, and it’s also, in many ways, destructive and bad for humanity for us to care about the well-being of these machines when we haven’t really figured out the well-being of humans. It also distracts from the real question of cybersecurity. Just in the past two or three days, we’ve had major platforms being hacked. We’ve had major programming languages like PHP having really core vulnerabilities. And if you’re not paying attention, if you’re running an enterprise and you’re not paying attention to this stuff, if you’re running a company on some of these platforms that are insecure, you’re putting your business and your users at risk.
Yeah, I don’t think the world has taken that seriously. It’s a disaster brewing. The Hugging Face incident was the first kind of shot on that, but it’s not the end. This week has been pretty brutal for cybersecurity.
MASAD: Our team is working nonstop. Overnight, everyone’s in the office making sure we’re mitigating whatever vulnerabilities are coming up. It’s a very serious moment for the world, and the talk about AI extinction is just not relevant right now. We need to focus on cybersecurity.
Copy LinkBuilding a moat around secure AI agents
SAFIAN: The fact that Amjad can refer to anything that’s happened at Replit in the past two years as slow and steady is kind of remarkable, and it underscores what’s so different about the pace of a true AI-native business. So is he at all worried that the big AI players will copy what Replit’s doing? And what does he see happening to human workers in an AI world? We’ll talk about that and more after the break. Stay with us.
[AD BREAK]
Before the break, Replit’s Amjad Masad talked about how Replit went from a $3 billion valuation to $9 billion in just six months. Now he talks about the competitive risks ahead posed by the biggest AI firms, how vibe coding is changing the work world, and more. Let’s jump back in. Let’s go to the last movie I have for you, number three, which is The Day After Tomorrow. Since it’s time to be looking forward, I will focus on the title, not the plot, although it did have a happy ending. I know you want to take Replit to a happy ending, but there are risks to be navigated. So I want to keep going down this road of the risks you’re talking about. The biggest tech players — Anthropic, OpenAI, Google, Microsoft, whatever — could build a version of what Replit is doing. Some of them are trying to. How do you create a moat that protects against that?
You know, what is your incentive? We were talking about this a little bit earlier, sort of the difference between your incentive and maybe their incentives.
MASAD: Yeah. So when you’re bringing AI into your organization, actually, Satya wrote a very good essay. He said AI agents are insider threats. The way I like to visualize it is: imagine inviting an employee into your office. That employee is very intelligent, very productive, but is slightly insane and can go off the rails and might be an actual security threat to you and the other employees. Do you rely on him or her for the security? Would you say, “Oh, go secure yourself”? Probably not.
You would want an external security company or an internal platform. So the way we think about Replit is it’s like the agent habitat. It is the place for agents to be productive, but also secure and governed. The way we architected our system is based on these zero-trust principles and security. We spent a lot of time working on the sandbox and the operating system underlying the sandbox. One interesting stat is there were about 100 zero-day vulnerabilities over the past year. Replit was affected by about zero of them because we actually built a special version of the kernel that is stripped down, so it has less surface area to attack.
We do a lot of work there. And when you read about the incidents of hacking coming out of the labs, they don’t have that expertise, nor are they incentivized to do that, because their incentive is to sell the greatest amount of tokens possible. In fact, the more problems they create, the more tokens they can sell because they can tell you, “Yeah, AI is creating all the security issues. By the way, here’s Mythos. It can help protect you.” So I think Replit and companies in the application space, our role is to make sure that AI is useful, safe, secure, and deployed in the rest of the economy in a secure, responsible way.
SAFIAN: And so I’m using those tokens in the most effective way.
MASAD: In the most effective, efficient way, and so on. I think the labs lack those kinds of incentives.
SAFIAN: Another sort of day-after-tomorrow topic would be that some people claim that vibe coding can break down when it’s deployed at scale. Is that a legitimate concern?
MASAD: It was, up until recently. I think we’re now at a point where you can say that vibe coding is not the most precise term, but let’s say agentic engineering is actually better than human engineering. The reason it’s like that is, for example, if you look at the kind of software we generate at Replit, increasingly we’re making it write in programming languages that humans kind of didn’t like and didn’t enjoy, like Rust.
Rust is a programming language that is memory-safe. What that means is that you can’t have null pointer exceptions or vulnerabilities coming out of memory overflow issues. So it’s a safe programming language, but it’s not been very ergonomic. Human engineers don’t like to write in programming languages like Rust. AI agents don’t mind, and so that’s one example where we can generate safer software and better software.
Another thing we’re doing is formal verification. When you’re working on something like the NASA space shuttle and you’re writing the software for that, oftentimes you might spend a year trying to formally verify it, meaning make mathematical proofs about the program you’re writing, and you would do it only in the most extreme life-or-death scenarios because it was very laborious, hard work. Now we’re having AI do formal proofs over random applications because it’s become very cheap. So I would say we’ve crossed the threshold to the point where automated software engineering is safer than human engineering.
Copy LinkHow AI could make work more human
SAFIAN: So do you have a vision of what the day after tomorrow looks like? In this extraordinary dynamic moment in history, you’re at the center of it. Can you see what the future is going to be?
MASAD: Yeah.
SAFIAN: Or is it too soon? Or is there a clear horizon?
MASAD: Well, I can sort of start to sense where it’s headed. We see Replit as customer zero of Replit, and the way we work at Replit has fundamentally changed starting earlier this year. A year ago, if I’m walking around our office, you’d hear keyboard clicking from engineers. They like these mechanical keyboards.
SAFIAN: Yes.
MASAD: They’re really annoying. But now I walk around the office, and it is actually very dynamic. It feels almost like a New York creative agency. Everyone’s on the whiteboard. People are arguing. People are talking. People are spending a little more time at lunch discussing what they want to build. People are spending more time with customers. It is actually, in many ways, more human.
I don’t think our place in the world is to sit down and act like machines. Our place is to interact with other humans, really spend a lot more time being creative, and think about what’s worth building and what’s meaningful. I see that a lot at Replit. Because we’re using agents to automate task-level work, we’re spending a lot more time on strategic thinking, and I think that is the most positive, bullish vision for the future. We started calling it the self-driving company. We wrote a blog post about how we see Replit as the first kind of self-driving company, and our mission has evolved from this company that’s making software more accessible. Well, it turns out that’s the detail. The second-order effect of making software more accessible is that we change the future of work. Work becomes a little more human. It becomes a little more creative, and we leave the mundane tasks to the machines.
SAFIAN: Well, I hope that vision is what we end up with. Thank you, Amjad.
MASAD: Thank you for having me. I appreciate it. Thank you.
Episode Takeaways
- Replit CEO Amjad Masad says the company’s breakout was years in the making, turning a long-shot belief that anyone could build software into the natural-language coding wave now called vibe coding.
- Masad recalls how close Replit came to the brink, with layoffs, stalled revenue, and a dispiriting office mood, even as a small internal war room was quietly building the agentic future.
- As rivals flooded into vibe coding, Replit resisted the hype chase and focused instead on reliability, security, and enterprise trust, winning serious customers building real businesses on the platform.
- Masad argues the real AI danger is not sci-fi talk about sentience or extinction, but a fast-rising cybersecurity crisis, and he says companies ignoring secure AI deployment are courting disaster.
- Looking ahead, Masad sees AI making work less mechanical and more human, with agents handling task-level coding so teams can spend more time debating, creating, and serving customers.