Cognition launches Devin Bots, coding agents that manage teams of Devins on projects for days or weeks

AI & Compute

Cognition Launches Devin Bots, Managers for Teams of Coding Agents

Cognition's Devin Bots manage teams of AI coding agents for weeks at a time. Usage grew tenfold in two weeks; the $48B startup passed $1B in annualized revenue.

By
Rebecca Stone
Filed
Channel
AI & Compute
Read
3 min read

Cognition, valued at $48 billion and past a $1 billion annualized revenue run rate as of September, has launched Devin Bots — AI agents that manage whole teams of Devin coding agents on software projects that run for days or weeks at a time.

CEO Scott Wu announced the product in an exclusive with Fortune, framing it as a bet that the software layer around AI models will matter more as the underlying models get smarter. "Self-driving software is, I think, the direction that it's going," Wu said.

The early signal is steep, if thin on detail: bot usage grew tenfold in two weeks among early-access customers, Fortune reported.

What do the bots actually do?

A user gives a bot a goal. The bot breaks it into tasks, spins up Devin sessions in parallel — each on its own cloud machine — and keeps them running until pull requests are ready to merge. "Your Devin Bot leads a team of agents to build, test, and improve your software around the clock," Cognition says on its product page.

Fortune's worked example: a bot asked to build an Android version of an iPhone app splits the job among Devin agents, then keeps the Android app updated as the iPhone app changes over time.

Cognition ships templates for specific roles:

  • A QA engineer bot
  • An Android engineer bot
  • A latency optimizer
  • A "chief of staff" bot that delegates work to other bots

The bots behave like teammates. They join a team's Slack, can be texted or called, watch tools such as Slack and Linear on schedules, and ping humans only when they're blocked or facing a decision that needs a person. Each bot carries its own budget, measured against the results it delivers, according to Fortune.

Wu said a bot picks models from several labs, routing routine code to cheaper ones. That flexibility is the strategic core of the product, and it leans on technology from The Interaction Company — maker of the Poke assistant — which Cognition acquired in July.

How does Cognition handle the risk?

Autonomy at this scale invites security questions. Bots run on separate cloud machines, isolated from keys sitting on engineers' laptops. Administrators decide which repositories and tools each bot can reach. Network controls aim to block unauthorized data transfers, including the case where a model follows malicious instructions hidden in content it reads — the classic prompt-injection scenario.

Teams also choose which changes require human review; low-risk edits can be approved and merged automatically. Fortune did not report pricing or a general availability date.

Can Cognition outflank its own suppliers?

Devin Bots land in a crowded lane. Anthropic's Duet works with Claude Code, and OpenAI's Dots runs teams of agents on software tasks through Codex. Both labs supply the models that power Devin — and both compete with it directly.

Fortune laid out the squeeze plainly: better models improve Devin, but a better Codex or Claude Code could give customers less reason to pay for Devin at all. Wu's answer is model independence. A bot isn't tied to one lab, he said — it chooses models for planning, design and computer use. Cognition also trains its own cheaper software-engineering and code-review models in-house.

The customer list suggests enterprise traction. Early-access users include Decagon, Fireworks AI and Reducto. Citigroup told Fortune that 14,000 of its engineers use Devin.

That number anchors the commercial picture: a $1 billion run rate, a $48 billion valuation, and a tenfold usage spike in two weeks all hinge on whether the orchestration layer — not the models — becomes the thing enterprises pay for. With Anthropic and OpenAI shipping rival always-on agents, that question gets answered in the market, not on a roadmap.

Source: mlq.ai

Share this article:

More from Rebecca Stone

Rebecca Stone

Show full bio

Correspondent covering technology and product news in Semiconductors, chip manufacturing, hardware, AI compute at Chip Dispatch.

1,092 articles

Related articles

« Previous articleNext article »