Just before Christmas in 2023, the small Cognition team was struggling to set up a particularly complex data server for San Francisco-based AI startup Devin’s fledgling coding assistant. They had spent hours studying installation documents and trying different commands, but couldn’t get it to work. Tired and frustrated, they decided to see how Devin would handle it.
When AI went into action, it baffled its creators, as it “executed the most witchy and black magic commands,” recalls co-founder and product manager Walden Yan, 21.
For a while, it seemed like Devin wouldn’t do any better than them. Then, a server terminal light that had been red for hours turned green. The data server was up and running.
They realized that Devin had deleted a faulty system file that the team had missed. “That was the moment I really realized how much software engineering is going to change,” Yan says.
It was the first major task Devin completed and proof of concept for Cognition’s vision of an AI that takes the heavy lifting out of coding. Now, almost a year later, Devin handles basic engineering tasks: detecting and fixing bugs, updating code fragments, and migrating them between platforms. Give it a simple instruction: “clean up this codebase,” and create an action plan and execute it. Most of the time, it works.
It’s a different approach from other, better-known and larger players in the still-burgeoning field, such as Github (which Microsoft bought for $7.5 billion in 2018) and Codeium, valued at $1.3 billion, which provide digital assistants that help people to write code with AI-powered suggestions.
But Devin is an autonomous AI agent that, in theory, writes code itself, without people involved, and can complete entire projects that are usually assigned to developers (the name Devin comes from “dev,” an abbreviation of the term ). “What we saw is a real opportunity,” says Scott Wu, 28, co-founder and CEO of Cognition, “to move from completing text to completing tasks.”
AI-generated code is already starting to change the industry. In October, Google CEO Sundar Pichai said that more than a quarter of the tech giant’s new code is written by AI. At Github, which hit an annual run rate of $2 billion in 2024, its code completion tool accounted for 40% of revenue growth this year, Microsoft CEO Satya Nadella said in July.
Pitchbook analyst Brendan Burke says AI coding has become the most funded use case in generative AI, with startups focused on it raising more than $1 billion in the first half of 2024 alone.
Actual revenues have only just begun to grow: research firm IDC expects them to surpass just $4 billion by 2029. Several AI coding startups surpassed $10 million in annual revenue run rates; Cognition declined to share the revenue, but president Russell Kaplan says dozens of clients signed up, with a typical six- to seven-figure annual contract.
But the opportunity is exciting enough that giants like Anthropic, Amazon and IBM have launched their own coding tools, along with startups like Poolside (valued at $3 billion) and Anysphere (valued at $400 million).
The ability to write code has already become a “safe bet” for the typical AI model, says IDC analyst Ritu Jyoti, who notes that the leader in the space remains OpenAI’s ChatGPT. But AI coding that can operate entirely on its own, like what Cognition is developing, “will bring about a sea change.”
We recommend: Numenta, backed by the Gates Foundation, launches an open source AI model
These are the companies that have invested so far in Cognition
That may not be great news for the 5 million Americans who work as programmers and earn a median salary of $130,000, let alone the 13 million programmers in India and China. Wu insists that massive job losses are not imminent and that the field has been “supply-constrained.”
Everyday programmers may be cautious, but investors love it. Peter Thiel’s Founders Fund and Khosla Ventures are betting on Wu and his 25-person team, investing $176 million in Cognition in a Series B round in April, raising its valuation to $2 billion just six months after its founding . The cash injection came just three months after the startup closed a $21 million Series A in January.
His clients include the $300 million expense management company Ramp, which uses him to write tests and clean up dead code, and the $1.7 billion data platform MongoDB, for which Devin updates the code architecture obsolete, saving its customers millions, says product manager Sahir Azam. Programmers at the $8 billion fintech Nubank, meanwhile, use it for tasks like updating code repositories.
It’s still early, but Founders Fund partner John Luttig invested in Cognition in part because he believes it has enough of a lead that “catching up with them in encryption agents is very difficult.”
Microsoft appears to have a similar hunch, having signed a partnership in May to offer Devin to developers on its Azure cloud, with chief technology officer Kevin Scott praising it as an “extraordinary” tool at the company’s annual developer conference.
But Silicon Valley is littered with companies that had a glorious advantage over Amazon and Google, until they didn’t.
Cognition’s promoters are betting that if anyone can create the ultimate coding machine, it’s three world-class programmers with the accolades to prove it. The founders are all Olympic-level gold medalist programmers who met on the competitive programming circuit. Wu was an elementary school math whiz with a “legendary grandmaster” ranking (the highest) on Codeforces, a networking site for programmers that hosts coding contests. Eric Glyman, a Cognition angel investor and CEO of Ramp, says Wu has one of the “top five IQs of anyone I’ve ever met.”
“Scott is clearly bright, curious, and boundless in ambition,” says Sarah Guo, a venture capitalist who invested in Cognition in three rounds of funding through her company, Conviction.
Wu is also a Forbes 30 Under 30 alum — he appeared on the list in 2019 for his previous company, Lunchclub, which uses artificial intelligence to set up networking meetings. A Cognition investor told Forbes that Wu left in 2022 because his interests had moved elsewhere, and Lunchclub, which has raised about $30 million, is moving forward.
Programmers assure that ‘Devin’ is not what they promise
Cognition launched Devin in March and generated a lot of buzz. In a demo video that racked up 30 million views on X, the company claimed that Devin had “successfully passed practical engineering interviews from leading AI companies” and completed complicated coding tasks. Some engineers were impressed by Devin’s technical skills; others feared for their jobs. Shortly after the release, a video of Wu as a seventh-grader dominating a math competition resurfaced, with online commenters joking, “He’s not human, he’s an AI” and “Devin is just Scott answering your questions on a messaging app.” ”.
Then, Carl Brown, a well-known independent developer from Austin, Texas, called out the shenanigans. In a video titled “Debunking Devin,” viewed more than 500,000 times, he accused Cognition of overselling its AI engineer. Their review found that Devin took much longer than a human would complete the job and that he introduced errors along the way.
Experiences like these led some to wonder if Devin is simply more inflated air in the AI bubble.
Ask it to design a flashy user interface and the results are drab, says Krish Manair, an engineer at data labeling company Labelbox, who tested the tool’s web application building capabilities.
Several rival founders told Forbes they believe the company promised too much about Devin’s capabilities, making it seem like an engineer could use Devin directly to code anything. Its current capabilities, they argue, are much more limited and geared toward predefined tasks like cleaning up existing code.
In a live demo, Forbes asked Devin to create an app for tuning a guitar. He created it in about 10 minutes, but the app couldn’t correctly identify the musical notes played to test it, and Cognition’s founders weren’t sure why.
“There is always this gap between the hype of what is possible and what works reliably,” says Varun Mohan, CEO of rival coding startup Codeium.
Wu admits that Devin is far from perfect. “Software engineering in the real world is very messy,” he says. “Humans write errors all the time.” And to be fair, multiple detractors were impressed with the tool’s capabilities.
Devin has improved significantly in the seven months since its launch, Wu says, thanks in part to feedback from early enterprise customers who seem to accept the agent as a work in progress. “We don’t see the tool as something that we hope is magical,” says Nubank CTO Vitor Olivier. “We are cautiously optimistic.” In some cases, he says, engineers with access to Devin were eight times faster at their jobs.
Earlier this year, it gave Devin the ability to cast subservient AIs to help him, but when Cognition tested this new “manager” mode, it discovered that sub-Devins would cast their own minions, who would in turn cast even more , creating a vast, endless loop of AI bureaucracy. “We finally had to cancel the job because they kept delegating work,” Kaplan says.
That said, Kaplan notes that the tool works best when multiple Devins work simultaneously on different projects, like an “army of young engineers.” That’s a phrase that likely makes some people who code for a living uncomfortable, and Wu says some of the negativity toward Cognition comes from anxiety about whether AI will eliminate software engineering jobs. Devin could allow companies to do more projects, he argues, by hiring humans to do the more meaningful work.
“There is really a lot of fear,” he says. “People have a lot of questions about what is happening in this new paradigm.”
This article was originally published in Forbes US
Subscribe to Forbes Mexico
