Cognition Bets $48 Billion Valuation on Devin Bots Coding Agents
Cognition, valued at $48 billion, launches Devin Bots to manage teams of AI coding agents, challenging OpenAI's Dots and Anthropic's Duet in a coding market where Cursor fetched $60 billion.
By Olivia Hart
5 min read

What's News
- Cognition, valued at $48 billion, launched Devin Bots, coding-specific manager agents for projects spanning days or weeks.
- Bot usage across dozens of early-access customers grew tenfold in two weeks, per Cognition.
- Cognition's annualized revenue run rate passed $1 billion in September.
- Citigroup's 14,000 engineers use Devin; CEO Jane Fraser said tech vendor spend fell 30% year-over-year.
- SpaceX acquired Cursor parent Anysphere in August at a $60 billion valuation.
Cognition, valued at $48 billion, has launched Devin Bots — persistent coding agents that manage teams of Devin agents on complex projects lasting days or even weeks. The launch is a direct challenge to OpenAI's Dots and Anthropic's Duet, two always-on agents built for general use.
The difference, Cognition argues, is focus. Devin Bots are among the first agents designed specifically for software development. Tell a Bot to build an Android version of your iPhone app, and it divides the work, assigns tasks to Devin agents and keeps the Android app updated as the iPhone version changes.
CEO Scott Wu frames the ambition plainly. "Self-driving software is, I think, the direction that it's going," he told Fortune at the company's office in San Francisco's SoMa neighborhood. "But of course, it's always up to the human where we're driving to."
What is at stake for Cognition?
Devin Bots is Wu's bet that as AI models get smarter, the product built around them matters more, not less. His position pushes against a broader Silicon Valley argument that advancing models will squeeze out companies building on top of them. Wu rejects the idea that AI will become so capable that, as he put it, "nothing else will matter at all."
The competitive pressure is real. When Cognition unveiled Devin in March 2024 — billed as the first AI software engineer — the launch video went viral. Two and a half years later, coding is one of AI's biggest businesses. In August, SpaceX acquired Anysphere, the parent of AI coding tool Cursor, in a deal valuing the company at $60 billion.
For Cognition, the model companies are both suppliers and competitors. Their advances make Devin more capable, but improvements to OpenAI's Codex and Anthropic's Claude Code could give customers less reason to pay for Devin at all.
How does a Devin Bot actually work?
Wu's argument is that capable models need more ways to be directed and inspected. GitHub Copilot began by suggesting code that programmers accepted or ignored. An engineer working with a Devin Bot might tell it what to optimize, how to use the team's project board, when to send updates, or ask for video of it testing its work.
He compares this to colleagues on a team. "Even here in this office, as you can see, we have whiteboards, we have scratch paper," he said. The more AI can do, the more ways people need to communicate with it.
According to the company, the specifics include:
- A Bot joins a team's Slack as its own member and remembers context across weeks.
- It contacts people when blocked or when it needs a decision.
- Several colleagues can work with the same Bot, including over iMessage.
- The product incorporates technology from the Interaction Company, the startup behind AI assistant Poke, which Cognition acquired in July.
Why claim an edge over OpenAI and Anthropic?
OpenAI and Anthropic are also building proactive agents that work across connected apps. Wu said Cognition's one advantage is that it is not tied to a single lab's models. A Bot selects suitable models for planning, design, computer use and other project components, using faster, cheaper ones for routine code.
"Developers will especially want to optimize cost as they use products like Devin Bots that do way more work than before," he said. Each Bot can be given its own budget, measured against the results it delivers. Cognition also trains its own models, which Wu said perform software engineering and code review at lower cost than frontier models. Devin combines these with outside labs' models.
Early traction numbers come from the company itself: usage of Bots across dozens of early-access customers — including Decagon, Fireworks AI and Reducto — grew tenfold in two weeks. A tenfold jump in two weeks is striking, but it starts from an early-access base the company did not quantify.
What are the numbers behind the business?
Cognition said its annualized revenue run rate passed $1 billion in September. Citigroup CEO Jane Fraser said on the bank's Oct. 13 earnings call that 14,000 of its engineers use Devin. "We've been in-sourcing work and have reduced our tech vendor spend by 30% year-over-year as a consequence," she said — roughly a three-tenths cut in one of the bank's supply line items.
How does Cognition handle the security risk?
Handing a Bot a project means deciding which systems it can access and which changes it can make without approval. Cognition runs Bots on separate cloud computers, keeping them away from passwords and keys stored on engineers' laptops, Wu said. Administrators set which code repositories and tools each Bot can access, and network controls aim to prevent unauthorized data transfers — including cases where a model follows malicious instructions hidden in content it reads.
Bots run their code through the same tests used for human-written code, Wu said. Teams decide which changes require human review and can configure low-risk edits, such as changing a button's color, for automatic approval and merging. The safeguards, he stressed, rely on rules companies set and monitor, "rather than relying on models themselves being smart enough to recognize a risky or malicious prompt."
That places the burden of oversight on customers, not on the technology itself.
Wu's stated aim is to keep making Devin more autonomous as models improve, positioning Cognition so that "we are happy when the frontier capabilities continue to improve, not sad." Whether product depth can outrun the model labs' own agents is the question the next year of this market will answer.
Source: fortune.com
More from Olivia Hart
Show full bio
Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.
2,301 articles