Field report W-17 / R7 · sheet /strategy
Corporate StrategyFiled from the territory
Cognition Bets $48 Billion Valuation on Devin Bots to Reclaim Coding Lead
Cognition, valued at $48 billion, launches Devin Bots — persistent agents managing multi-week coding projects — challenging OpenAI's Dots and Anthropic's Duet in the AI coding race.
- Grid
- W-17 / R7
- Surveyed
- November 18, 2026
- Reading
- 5 min read
- Correspondent
- Sophie Lindqvist

Field notes
- Cognition, valued at $48 billion, launched Devin Bots on the week of Nov. 17, 2026
- Bot usage across early-access customers grew tenfold in two weeks, the company said
- Annualized revenue run rate passed $1 billion in September
- Devin is used by 14,000 Citigroup engineers; the bank cut tech vendor spend 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 instances across projects lasting days or even weeks. The move is a direct challenge to OpenAI's Dots and Anthropic's Duet, the general-purpose always-on agents both labs rolled out in recent months. Cognition claims its Bots are among the first such agents designed specifically for software development.
CEO Scott Wu frames the product as the logical next step for the category his company helped create. "Self-driving software is, I think, the direction that it's going," he told Fortune at Cognition's office in San Francisco's SoMa neighborhood. "But of course, it's always up to the human where we're driving to." The caveat is doing real work here: autonomy, yes, but human-set destination.
What does Devin Bots actually change?
In practical terms, a Bot operates as a project manager for AI labor. Tell it to build an Android version of an iPhone app, and it divides the work, assigns tasks to Devin agents (autonomous AI programs that write and ship code), and keeps the Android build synchronized as the iOS version changes. Each Bot joins a team's Slack as its own member, retains context across weeks, and contacts people when blocked or when it needs a decision, according to the company. Several colleagues can work with the same Bot, including over iMessage.
The product incorporates technology from the Interaction Company, the startup behind the AI assistant Poke, which Cognition acquired in July.
Wu's argument cuts against a persistent Silicon Valley worry: that as models improve, companies building products on top of them get squeezed out. He rejects the notion that AI will eventually become so capable that, in his words, "nothing else will matter at all." His counter is that more capable models demand more ways to direct and inspect them. An engineer working with a Bot might specify what to optimize, how to use the team's project board, when to send updates — or request a video of the Bot testing its own work.
He compares this to ordinary teamwork. "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 channels people need to communicate with it.
A strained position: supplier and competitor
Cognition's predicament deserves scrutiny it rarely receives. OpenAI and Anthropic are both its suppliers and its rivals. Model advances make Devin more capable; the same advances improve Codex (OpenAI's coding tool) and Claude Code, giving customers less reason to pay for Devin at all.
Wu's answer is model independence. 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 assigned its own budget, measured against results. Cognition also trains its own models, which Wu said perform software engineering and code review at lower cost than frontier models (the most capable AI systems available).
The market context is unforgiving. When Cognition unveiled Devin in March 2024, calling it 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, parent of the coding tool Cursor, in a transaction valuing the company at $60 billion — well above Cognition's own valuation.
What do the numbers show?
Cognition said usage of Bots across dozens of early-access customers — including Decagon, Fireworks AI and Reducto — grew tenfold in two weeks. The figure is self-reported, covers an early-access cohort, and comes without absolute numbers; a tenfold rise from a small base is a common launch metric. The company also announced its annualized revenue run rate (a projection of yearly revenue from current performance) passed $1 billion in September.
The strongest independent datapoint comes from Citigroup. Devin is used by 14,000 engineers at the bank, CEO Jane Fraser said on the Oct. 13 earnings call. "We've been in-sourcing work and have reduced our tech vendor spend by 30% year-over-year as a consequence," she said. Notably, that efficiency gain translated into reduced spending elsewhere — a reminder that AI productivity gains often shrink someone's budget.
What about security and review?
Handing an agent a multi-week project raises the question of what it may access and change. 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 reach, 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 applied to 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. He was candid about the limits: the safeguards rely on rules companies set and monitor, "rather than relying on models themselves being smart enough to recognize a risky or malicious prompt." That is a policy burden shifted onto customers, whether or not they staff for it.
Wu's stated aim is to keep making Devin more autonomous as models improve, positioning Cognition so that, in his words, "we are happy when the frontier capabilities continue to improve, not sad." Whether the frontier labs let him stay happy is the question the next two years will answer.
via fortune.com/feed/ (Source)
More from Sophie Lindqvist
Show full bio
Correspondent covering money, funding and market shifts around corporate strategy and chief-executive leadership at CEO Frontier.
479 articles
Adjacent sheets
- Databricks launches always-on AI agents in enterprise push
- McKinsey and Cognition Team Up to Prove AI Coding Pays
- OpenAI Disbands Robotics Team as GPT-6 Astra Starts Running Robots
- Notion Bets $11 Billion on Taste: Ivan Zhao's Refounding Play
- Nearly 60% of Software Engineer Job Ads Now Demand Communication Skills