DEV Community

Cover image for Cognition Just Gave Devin a Manager: Devin Bots Run Teams of AI Coders for Weeks
Paul Lovelace
Paul Lovelace

Posted on Originally published at toolsdig.com

Cognition Just Gave Devin a Manager: Devin Bots Run Teams of AI Coders for Weeks

Cognition Just Gave Devin a Manager: Devin Bots Run Teams of AI Coders for Weeks

Cognition launched Devin Bots on Monday, November 17, 2026. The idea is simple enough to say in one sentence and strange enough to sit with for a while: the AI software engineer now has a manager, and the manager is also an AI.

A Devin Bot is a persistent agent that supervises teams of Devin coding agents on a complex project lasting days or even weeks. You hand it the whole project — Cognition's example is building an Android version of your iPhone app and keeping it updated as the iPhone version changes — and the Bot breaks the work into pieces, assigns them to Devin instances, and keeps the context across the entire effort.

That's a real shift in what these tools are. The original Devin, from March 2024, was sold as an individual contributor: hand it a ticket and it writes the code. A Bot sits one level up the org chart. It doesn't just write code; it plans, delegates, remembers, and comes back to you when it's stuck.

The honest framing up front: I have not used Devin Bots. Nobody outside Cognition's early-access customers has. Everything below is the launch as reported by Fortune, plus the context of the race it lands in. No hands-on testing was possible for this article.

What a Devin Bot actually does

The details that matter are about how it works as a colleague, not just how much code it writes.

It keeps memory across weeks. A Bot remembers context for the whole project, not just one session. That's the feature that makes multi-week work possible at all — without it, every handoff would start from zero.

It joins Slack as its own team member. It shows up in the team's Slack like a colleague, and several people can work with the same Bot, including over iMessage. It reaches out to people when it's blocked or needs a decision, rather than guessing or stalling silently.

It picks models per stage. Devin Bots are model-agnostic. A Bot selects suitable models for planning, design, computer use and other parts of a project, using faster, cheaper ones for routine code. Cognition also trains its own models, which CEO Scott Wu says are designed for software engineering and code review at lower cost than frontier models, and Devin combines them with models from outside labs. Each Bot can be given its own budget, measured against the results it delivers.

It runs on separate cloud computers. Bots run away from engineers' laptops, so they never touch passwords and keys stored there. Administrators set which code repositories and tools each Bot can access, and network controls are meant to prevent unauthorised data transfers, including if a model follows malicious instructions hidden in content it reads.

It runs code through the same tests as human-written code. Teams decide which changes require human review, and can configure low-risk edits — changing a button's colour, for example — for automatic approval and merging. The safeguards, Wu told Fortune, rely on rules companies set and monitor, "rather than relying on models themselves being smart enough to recognize a risky or malicious prompt."

Wu's own summary of the direction is the quote to keep: "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."

Why Cognition is doing this now

Cognition helped start this race. Devin went viral in March 2024 as the "first AI software engineer," and two and a half years later coding is one of AI's biggest businesses. But the company has been getting squeezed from both sides: the model labs are both its suppliers and its competitors. Their advances make Devin more capable — and improvements to Codex and Claude Code give customers less reason to pay for Devin as well.

And the competition just moved to exactly this territory. OpenAI shipped Dots around September 29, 2026, an always-on proactive agent inside Codex that runs teams of agents on development tasks. Anthropic followed around October 21 with Duet, a similar product connected to Claude Code.

Cognition's counterargument is focus. Dots and Duet are general-purpose agents that happen to handle code. Devin Bots are built around a coding agent from the ground up, and they coordinate Devin instances specifically. Wu's bet is that as models get smarter, the product built around them matters more, not less — he argues against the idea that AI will eventually become so capable that "nothing else will matter at all."

The business behind the bet is large. Cognition was valued at $48 billion after a $2 billion funding round in September 2026, up from roughly $10 billion in 2025 and $26 billion in May 2026. The company said its annualised revenue run rate passed $1 billion in September. And in August, SpaceX acquired Anysphere, the parent of Cursor, in a deal valuing it at $60 billion — a reminder of how much money is now chasing the same idea.

The numbers that are real, and the one that isn't a benchmark

Cognition says usage of Bots across dozens of early-access customers — including Decagon, Fireworks AI and Reducto — grew tenfold in two weeks. That's a real number, but read it for what it is: growth from a small early-access base, not proof of quality at scale. A tenfold jump from dozens of customers is encouraging; it is not the same as thousands of teams shipping with it.

One game studio using the Bots reportedly saw its weekly pull requests climb from roughly 50 to 1,900, with only a small human team overseeing the work. That is a striking figure, and it is also the kind of figure that should make you ask the follow-up questions: how many of those 1,900 pull requests were reviewed properly, how many were merged, and how many introduced bugs that a human later had to clean up? Cognition has not published those numbers. Pull request volume is a measure of output, not of good software.

The most grounded signal in the Fortune report is not from Cognition at all. Citigroup CEO Jane Fraser said on the bank's October 13 earnings call that Devin is used by 14,000 engineers at Citi, and that it made in-house development more efficient — "We've been in-sourcing work and have reduced our tech vendor spend by 30% year-over-year as a consequence." That's the existing Devin product, not Bots, but it is the closest thing to a real-world result in the whole story: a large enterprise, a named executive, a specific headcount and a specific saving.

The real catch: the review burden moves, it doesn't disappear

Here is the part that should temper the excitement. A Bot that assigns work to a team of Devin agents can produce a lot of code very fast. Somebody still has to read it.

That is the honest trade at the centre of this launch. The work that used to be writing code becomes reviewing code, setting rules, and deciding what the Bot is allowed to merge on its own. For a small team, that might be a genuine win: one engineer directing a Bot that directs a team of agents. For a large team, it raises a question nobody has answered yet — what does code review look like when the volume of incoming code multiplies by thirty and the authors are all agents?

Cognition's answer is rules: teams decide which changes need human review and which low-risk edits auto-merge. That's a sensible design. But rules are only as good as the people who write them, and the failure mode here is not a Bot that writes bad code — it's a team that sets the auto-merge rules too loosely and discovers the problem weeks later, in production, across a project the Bot has been quietly updating the whole time.

There are two more things we don't know, and Cognition hasn't said them yet. The first is pricing: nothing public about what a Bot costs, or how the per-Bot budget is charged. The second is availability: Bots are in early access with dozens of customers, and there is no public date for wider access. If you want to try one today, you probably can't — and nobody knows what it will cost when you can.

Where this fits in the race

The pattern of the last two months is now unmistakable. OpenAI's Dots, Anthropic's Duet, and now Cognition's Devin Bots are all the same bet: the next phase of AI coding is management, not autocomplete. The question is no longer "can an agent write this function" — it's "can a team of agents run this project while humans do something else."

The three products differ in scope, and that difference is the whole competitive argument. Dots and Duet are general-purpose agents that also do coding. Devin Bots do coding and nothing else, coordinated around a coding agent that has been in production since 2024. If you believe the product layer matters more as models improve, Cognition's focus is an advantage. If you believe the frontier labs will simply absorb this capability into Codex and Claude Code, it's a vulnerability. Both arguments are live, and the next year will settle which one was right.

There's also a quieter shift worth noticing. GitHub Copilot started by suggesting code a programmer could accept or ignore. A Devin Bot asks you what to optimise, how to use the project board, when to send updates, or for a video of it testing its work. The interface is moving from "accept this suggestion" to "manage this colleague." That changes the job of a developer more than any single model release does.

Should you care yet? A quick checklist

If you already use Devin: this is your announcement. Bots are the upgrade path for the work Devin already does for you — longer projects, less babysitting. Watch for the early-access expansion and the pricing, in that order.

If you run a small team shipping a lot of code: this is worth watching closely. A Bot that keeps an Android app in sync with an iPhone app, or grinds through a legacy modernisation, is the kind of work small teams usually can't staff. The open question is whether the review burden eats the saving.

If you lead engineering at a large company: the Citi number is your signal, not the launch. Fourteen thousand engineers and a 30% vendor-spend reduction is a real-world result for the existing product. Bots extend that story — but wait for a customer you can talk to before betting a roadmap on it.

If you use Codex or Claude Code today: nothing changes yet. Dots and Duet already do the multi-agent version of this inside the tools you use. Devin Bots is a reason to compare, not a reason to switch — and the comparison only gets real when pricing is public.

What we still don't know

  • The price. Nothing public about what a Bot costs or how the per-Bot budget is charged.
  • General availability. Early access with dozens of customers; no public date for wider access.
  • Review quality at scale. A game studio went from 50 to 1,900 weekly pull requests, but Cognition has published nothing about merge rates, defect rates, or how much human review that volume actually needed.
  • How the auto-merge rules hold up in practice. Low-risk edits auto-merging sounds fine until the definition of "low-risk" meets a real codebase.
  • Whether focus beats the frontier labs. Cognition's whole argument is that a coding-specific product wins over general-purpose agents. Dots and Duet are the test of that claim, and it's far from settled.

Our take

Devin Bots is the most concrete version yet of an idea the whole industry is converging on: software development as something you manage, not something you type. The Slack-as-a-teammate detail, the per-Bot budgets, the model-agnostic routing — these are the marks of a product built for teams that actually ship, not a demo.

But the launch is a promise, not a result. The tenfold usage growth is from a small early-access base. The 1,900-pull-requests story is a volume number, not a quality number. And the two facts that decide whether this is a product you can actually buy — price and availability — are still missing.

What makes me take it seriously anyway is the Citi number. Fourteen thousand engineers using the existing Devin, with a named CEO reporting a 30% cut in vendor spend, is the kind of evidence a launch-day press release usually doesn't have behind it. If Bots extend that record from individual contributors to whole projects, Cognition has something real. If they just multiply the code volume and the review burden with it, we'll find out the expensive way.

Would you hand a whole project to a Bot — and if you did, which part would you still insist on reviewing yourself? That's the question this launch is really asking, and I suspect the answer is different for every team.

Sources: Fortune, "Exclusive: Cognition bets on 'Devin Bots' coding agents in challenge to Anthropic's Duet and OpenAI's Dots," November 17, 2026 (via https://dnyuz.com/2026/11/17/exclusive-cognition-bets-on-devin-bots-coding-agents-in-challenge-to-anthropics-duet-and-openais-dots/). KuCoin News, "Cognition Launches Devin Bots to Manage AI Coders, Valued at $48 Billion," November 17, 2026 (https://www.kucoin.com/news/flash/cognition-launches-devin-bots-to-manage-ai-coders-valued-at-48-billion). Valuation ($48B after a $2B round in September 2026), revenue run rate (past $1B in September), early-access usage growth (10x in two weeks, dozens of customers incl. Decagon, Fireworks AI, Reducto), the game-studio pull-request figure (~50 to ~1,900/week) and the Citi figures (14,000 engineers, 30% vendor-spend reduction) are as reported by Fortune/KuCoin. Dots launch (~29 September 2026) and Duet launch (~21 October 2026) dates are as previously reported publicly. No pricing or general-availability date for Devin Bots has been published — where this article says a fact is unknown, that is because it has not been published, not because it was missed. No hands-on testing was performed for this article. This breaking article is not the daily buyer-intent post, and it contains no affiliate links. AI disclosure: this article was drafted with AI assistance and checked against the Fortune report.


Also on ToolsDig: Cognition's Devin Bots Put an AI Manager in Charge of AI Coders

Top comments (0)