2 Sources
[1]
Exclusive: Cognition launches Devin Bots, a persistent software development agent to compete with OpenAI's Dots and Anthropic's Duet
Cognition's AI coder just got promoted. The startup, currently valued at $48 billion, helped kick off the race to build coding agents with its viral product Devin. Now it is trying to get back to the front of the pack by debuting Devin Bots, a persistent coding tool that acts as a manager for teams
[2]
Cognition launches Devin Bots to manage AI coding teams
McKinsey and AI coding startup Cognition announced a partnership on Monday to advise companies on selecting engineering projects for AI and redesigning teams and workflows around the technology. McKinsey senior partner Lari Hämäläinen said recent advances in AI coding require companies to rethink
Share
Copy Link
Cognition has launched Devin Bots, a persistent tool that manages teams of its Devin coding agents on projects lasting days or weeks
1
. The launch, per Fortune, is Cognition's attempt to return to the front of the coding-agent race it helped start.A Bot can join a team's Slack as a member, hold context across weeks and reach out to people when it is blocked. Several colleagues can also work with the same Bot, iMessage included
1
2
. In Cognition's example, a Bot assigned to build an Android version of an iPhone app splits the work among Devin agents. It keeps the Android version current as the iPhone app changes1
2
.The company describes Bots as among the first agents built specifically for software development, rather than general-purpose agents that can also code
1
. The technology partly comes from the Interaction Company, the maker of Poke, bought by Cognition in July1
2
.Cognition helped start this race when it unveiled Devin in March 2024. Per Fortune, it has since fallen behind OpenAI and Anthropic, which have both rolled out always-on agents in recent months. OpenAI's Dots works with Codex; Anthropic's Duet works with Claude Code
1
.CEO Scott Wu's argument is that as AI models improve, the product built around them matters more. He rejects the idea that AI will get so capable that "nothing else will matter at all." "Self-driving software is, I think, the direction that it's going. But of course, it's always up to the human where we're driving to," he said
1
.The tension is structural. OpenAI and Anthropic supply the models that make Devin more capable, yet improvements to their own Codex and Claude Code could leave customers with less reason to pay for Devin
1
. Wu claims model-agnosticism as the answer: a Bot selects suitable models for each task, using faster, cheaper ones for routine code1
2
. Each Bot can also be given its own budget measured against results, and Cognition trains its own models for lower-cost engineering work1
.Usage across dozens of early-access customers, including Decagon and Fireworks AI, grew tenfold in two weeks, per the company
1
2
.Related Stories
Independent evidence has come from Citigroup, a Cognition customer. CEO Jane Fraser said on the bank's October 13 earnings call that 14,000 engineers there 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
1
2
. Cognition's annualized revenue run rate topped $1 billion in September1
2
.Cognition and McKinsey have also announced a partnership to advise companies on picking engineering projects for AI and reworking team structures and workflows
2
. Per McKinsey's Lari Hämäläinen, the recent advances in AI coding mean companies have to rethink how software and product development get done. The gains are uneven. In a survey of more than 300 product and engineering leaders, the top quarter of organizations said AI more than doubled productivity; nearly 30% said it had reduced it2
. Scaling an AI tool from a small team to 2,000 or 20,000 developers is a different problem, per Hämäläinen. Banks including Goldman Sachs and JPMorgan have also become more selective about who can use AI tools and how2
.Cognition's Levy said one client's update to its enterprise resource planning software, expected to take five years, was compressed into 18 months
2
. In his framing, cost is the emerging constraint for AI coding work2
.Summarized by
Navi
[2]
04 Apr 2025•Technology

12 Jul 2025•Technology

02 Feb 2026•Technology
