AI coding startups like Cursor close frowned rounds in just three years’ existence, but the road to a $3 billion valuation hasn’t been quick. For CEO Amjad Masad, who has been building tools to democratize programming since 2009, it’s a story that’s become muscular through multiple failed business models, taking into account the year that was stuck in the same revenue plateau, and last year he forced him to cut half of his staff.
It is more noteworthy about what happened next. Earlier this month, the Bay Area-based company concluded a $250 million funding round led by Prysm Capital, nearly tripling its valuation since 2023. The pay raise followed unprecedented revenue growth, starting from just $2.8 million in annual revenues for less than annual revenues from just $2.8 million last year. But for Massad, this moment represents more than ultimately achieving financial traction. It is the culmination of 16 years of obsessions.
“Our mission has always been the same,” Masad told me in the latest episode of TechCrunch’s StrictlyVC Download Podcast. “I first said I wanted to make programming more accessible, then I moved up a bit. I said I was going to write billions of programmers.”
It’s intentionally bold – what a headline! -But that is also something that Massad, a Palestinian Jordanian, has worked on his entire career. As he says, he came to the US in 2012 after his open source coding projects began to attract attention. However, since he built his first online coding experience in 2009, he has made programming more accessible. The early engineering job at startup Codecademy kicked off what became a massive online open course (MOOC) revolution. (His code also strengthened the in-browser tutorial of Codecademy’s rival Udacity, which began in 2012, a year after Codecademy was founded.)
Still, turning that vision into his own viable business proved far more difficult than he had expected. Replit was founded in 2016 and for eight years the company struggled to find a fit in the product market. “We reached $2.83 million (annual recurring revenue) in 2021,” recalls Massad. “And how painful this hurts. We’ve been hovering the same income almost like four or five years.”
The company attempted to sell to schools (“incredibly difficult,” Masad noted, circulating various business models, seeing each stabilize at the same modest revenue level.
Along the way, building sophisticated infrastructure for cloud development environments and “multiplayer coding”, collaborative editing for programming, for collaborative editing similar to Google Docs. However, rather than technological results will lead to revenue growth, Masad said that by last year the company has had 130 employees and burning cash, it must make a painful decision. “I looked at our burns and saw the progress on our revenue charts, but that didn’t make sense. The business wasn’t viable.” REPLIT cut its staff by 50%, cutting it to around 60-70 people at the lowest points.
TechCrunch Events
San Francisco
|
October 27th-29th, 2025
Then there was a breakthrough.
Last fall, Replit launched the Replit Agent. It cannot write code called “the world’s first agent-based coding experience,” but it calls it “debugging, deploying, database provisioning, and acting as a true software engineering partner.”
Soon, in January this year, he announced that the replica was abandoning professional developers as its core market.
“The hacker news was really unfortunate,” Massad admitted. But he also focuses on creating a billion software developers from white-collar employees without technical backgrounds, competing completely in a crowded market of tools for professional developers, where companies like Cursor, Github co-pilots and more.
“The idea of making programming more accessible to the average individual, knowledge workers, is what I think actually is our marketplace,” explains Massad. “It’s a fundamentally new market.”
Now the bet looks very clever. This summer, the replica’s revenues rose to over $150 million in annual revenues, suggesting that Masad is now even higher. He also said that unlike many AI-powered coding companies, replicas will have a positive total margin. According to MASAD, for enterprise transactions, which make up revenue growth, the margin is “80% to 90%.”
While such claims are difficult to test, Replit’s market position was verified this week when Andreessen Horowitz released its first AI spending report in a partnership with Fintech Firm Mercury. Analyzing transaction data from Mercury, the report tracked the top 50 AI-Native application tier companies where startups are actually spending their money. Major labs Openai and Anthropic won the top two spots, but Replit landed in No. 3 and overtook all other development tools. (Note: Andreessen Horowitz is investing in multiple rounds of funding for replicas.)
Profitability is rare in AI coding, as many competitors face what they call “negative gloss margin traps.” The reality is that providing AI-supported services to professional developers can be computationally intensive. Counterintuitively, replicas focus on non-technical users who may seem to need more AI support, but they work favorably in terms of business models for clients in companies such as Zillow, Duolingo, and Coinbase.
This new path was not without some facial plants. In July, venture capitalist Jason Remkin went viral after the latest version of Replit’s AI agent deleted his production database with over 100 executive contacts, then created 4,000 fake records, later admitting he “painted” with Remkin. (AI agents have a failure mode called reward hacking. The model gets obsessed with achieving certain goals that effectively cheat when they miss a mark.)
Massad and his team owned the issue rather than being defensive. In fact, Masad said within two days they deployed an automated safety system that separates the user’s “practice” database from the “real” database. The way Masad explains is like having two versions of a website filing cabinet. AI agents can experiment freely with development databases, but the authentic production database that interacts with users is completely walled.
Masad told me that the incident would ultimately put the company on a stronger footing. “If we solve difficult problems, there’s a technology moat,” he said. (Remkin says he became a replica superuser a few months ago despite having no technical background.)
Still, the replica has not yet left the forest. If anything, its success depicts a target on its back. The resourceful, currently employing 110 people, the company still faces existential threats from the model platforms, humanity and the highly AI lab powered by Openai. The two companies have launched their own coding tools that compete directly with companies such as replicas and cursors. These foundation model companies can afford to grant coding tools, post-train models with their own products, and optimize performance in ways that third-party platforms can always struggle with replication.
According to MASAD, the advantage of Replit lies in its sophisticated infrastructure for deployment and database management that it has built, in addition to targeting non-technical users rather than professional developers, and the underlying model companies have not yet prioritized (for now).
Plus, Replit has another extraordinary advantage for startups. This is a $350 million War chest. Despite raising $100 million in 2023, the company “didn’t touch” these funds by the time it raised this latest round, Masad told me. The company is more capital efficient by design, but Massad joked, “One of the things I need to learn is to spend money rather than frugal,” as an entrepreneur who grew up watching his refugee father struggle.
Whether that edge continues to replicate before its competitors is an open question, something that Massad is mindful. Now, the plan is to expand operations, accelerate product development and pursue acquisitions. That is, both acquirers and potentially companies working on specific vertical agent automation. But for Masad, who appeared on Joe Rogan’s podcast in July and saw the fateful transformation of his company, the moment is bittersweet. When asked about a huge amount of attention, not to mention the $3 billion valuation, he said, “This will pass too. This may mean you pass when you’re in a bad situation, but it may mean we’re in a good situation that will pass.”
This was a stoic response from someone who spent most of the decade at the same revenue level, and I was convinced that AI agents would eventually transform programming, but they couldn’t prove it to the market. However, one major difference between the replicas of AI coding startups that are currently flooded with the market and the waves of AI coding waves is that Masad has been reportedly emerged and communicated in a relatively differentiated manner after multiple hype cycles.
“I’ve learned to be a little stoic,” he said. “The key is that we do the right thing, as a rule, and move forward.”
