LATEST
AI Tools

Devin Bots: The AI Manager That Runs a Team of Coding Agents

Updated November 19, 2026
15 min read
Editor's Note: This article was human-reviewed for accuracy
Team of software developers working together on computers in a modern tech office

Quick Answer & Key Takeaways

  • • Cognition launched Devin Bots on November 17, 2026, persistent agents that own an engineering goal and direct teams of Devin coding agents to deliver it over days or weeks.
  • • Bots are included only on Devin's $200-a-month Max plan and the Teams plan, so there is no cheap way in and no standalone Bots price.
  • • Cognition's launch numbers are striking, including one Bot merging 1,134 pull requests in 44 days, but every figure is company-reported and no independent test exists yet.

Devin has always been the coder you hand a ticket to. On November 17, 2026, Cognition gave it a manager. Devin Bots are persistent agents that take ownership of a whole engineering goal, keep Android at parity with iOS, keep a web app fast, or run a game studio, and direct teams of Devin coding agents to deliver it over days or weeks, coming back to a human only when a decision needs judgment.

This is the third entry in a race that started two months ago. OpenAI shipped Dots in late September 2026, Anthropic followed with Duet in late October, and Cognition has now answered with a product built around the coding agent it has sold since Devin went viral in March 2024 as what the company billed as the first AI software engineer. The launch numbers are loud: one Bot merged 1,134 pull requests in 44 days. Every one of those numbers comes from Cognition itself. This article works through what Bots do, what they cost, and the parts Cognition has not answered yet.

What a Devin Bot actually is

Cognition describes a Bot as "a persistent agent with its own memory and tools that can lead multiple Devins in parallel." The shift is from task to responsibility. Until now, working with a coding agent meant delegating specific work: the agent handled the implementation while the engineer stayed responsible for coordination, noticing the next problem, choosing an approach, writing the task, checking the result, and deciding what comes next. A Bot takes on the responsibility instead, investigates, delegates work to Devin sessions, and evaluates the results to decide what to pursue next.

A task to completeA responsibility to own
Port this feature to Android.Keep Android at parity with iOS.
Fix the layout shift.Keep the web app smooth as it changes.
Fix this onboarding error.Get new users through setup successfully.

The examples in that table come from Cognition's launch post, and they show the idea clearly: a task describes the change you want, while a responsibility describes the outcome and leaves the Bot to work out how to reach it. A Bot that owns a goal can also find work nobody would have assigned, because it watches everything that affects the goal, including the parts no one is looking at.

Six starter templates ship on the product page: Software Factory, Android Engineer, Chief of Staff, Latency Optimizer, Product Engineer, and QA Engineer. The Software Factory template finds bugs, polish, performance, and refactoring work in a product area you name and opens one small pull request for each, with proof that it works, learning from what you merge and close.

The three jobs Cognition gave its own Bots

Cognition spent the weeks before launch running Bots on its own products. The three examples in its launch post are the whole case for the product so far, so they are worth reading closely, including what they do not prove.

1. Building Devin's Android app

Devin has an iOS app and a team of iOS engineers behind it. Cognition had no Android engineers and no plan for Android. On September 28, 2026, an engineer on the Bots team created "Android Bot" and gave it one instruction: product-manage the Android app, build the whole thing autonomously, and work out the blockers first. Forty-four days later, Devin had an Android app, now in open testing on Google Play, and the Bot owns its development end to end, updating it every time a feature lands on iOS.

The numbers Cognition reports for those 44 days are: 1,134 pull requests merged, 344,651 lines of Kotlin written, 2,841 screenshot tests added, and 1,217 tickets closed. The engineer sent the Bot 183 messages, about one for every six pull requests it merged, and sent none at all on 22 of those days. Most of his messages were decisions the Bot needed a human for, such as a pull request that needed a security review. The Bot filed tickets itself, started a Devin session for each one, merged the pull requests that passed CI and Devin Review, and posted a daily report in Slack with a screen recording of what had changed.

2. Keeping the Devin app from jumping

The second example is smaller and, in some ways, more convincing. The Devin web app loads different parts of a page through separate requests, and when content arrives late it can push elements around on screen. These layout shifts depend on request timing and connection speed, so they are hard to catch in code review, and fixing one does not stop a new component from introducing another.

Cognition gave a "Layout Bot" the job of keeping the web app smooth. Every three hours it builds the production version of the latest code, runs the full stack on its own machine, and simulates production loading, including first visits, repeat visits, and loads with individual requests deliberately delayed. When a check fails, it reruns it to rule out timing noise, traces the regression to the pull request that introduced it, and opens a fix with before-and-after recordings. In three weeks it merged 159 pull requests and cut interface lag by 70%, according to Cognition. At Cognition, the company says it merges more than 2,000 pull requests a week, so this Bot owns a real, recurring slice of the work.

3. Running a game studio with one human

The third example is the most extreme. Jenny Xu runs a mobile game studio, and Cognition proposed a challenge: run the studio entirely on Devin Bots. Today she is the studio's only human employee, with about 20 Bots, each owning one job end to end: development, Unity, releases, UI, iOS, Android, and testing on real devices. Above them sits a "Super Jenny" Bot, the only one she talks to. Four times a day it collects the decisions the other Bots need from her into one Slack thread, and she answers with emoji.

The studio went from about 50 pull requests a week to 1,900, and ships a release every week, according to Cognition. Xu has since joined Cognition full time. The same pattern shows up in a customer quote on the product page: Bilt's VP of Engineering says its Devin Bots merged about 1,600 pull requests in a week. These are volume figures, and volume is a measure of output, not of good software. Cognition has published nothing about merge rates, defect rates, or how much human review that volume needed.

How a Bot works day to day

A Bot keeps memory across sessions, which is the feature that makes multi-week work possible at all. Without it, every handoff would start from zero. It joins a team's Slack as its own member, and several people can work with the same Bot, including over iMessage and in the Devin app. It reaches out when it is blocked or needs a decision, rather than guessing or stalling silently. Decisions that need a person land in a "Needs You" inbox, and Bots share reports and test recordings proactively, adapting to how you prefer to review evidence.

Bots are model-agnostic. A Bot picks different models for planning, design, and routine code, using faster, cheaper ones where it can. Cognition also trains its own models, and CEO Scott Wu told Fortune they are designed for software engineering and code review at lower cost than frontier models. Each Bot can be given its own budget, measured against the results it delivers. As you add more Bots, they coordinate with one another: a QA Bot finds a bug and brings in a Developer Bot to fix it, then tests the result, while a PM Bot coordinates everyone in Slack.

On safety, Bots run on separate cloud computers, away from passwords and keys stored on engineers' laptops. Administrators decide which repositories and tools each Bot can reach, and network controls are meant to prevent unauthorised data transfers, including if a model follows malicious instructions hidden in content it reads. Code passes the same tests used for human-written changes. Teams decide which changes need human review and can set low-risk edits, such as a button colour change, to merge automatically. 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." The product page lists SOC 2, ISO 27001, and FedRAMP marks, zero data retention with model providers, audit logs, and a separate virtual machine for every session.

What it costs

Here is the part most launch coverage missed: there is a price for Devin Bots, just not a standalone one. Bots access is a feature of two plans on Devin's pricing page, read on November 19, 2026. The Pro plan at $20 a month does not include it. To get Bots as an individual, the entry point is Max at $200 a month. For teams, Bots come with the Teams plan.

PlanPriceDevin BotsWhat else it adds
Free$0 (£0 / CA$0)NoLight quota to code with agents, limited model availability, unlimited inline edits and tab completions
Pro$20 a month (£15.20 / CA$28.65)NoFrontier models from OpenAI, Claude, Gemini, and SpaceX AI, free use of SWE-2, access to Devin Cloud, extra usage at API pricing
Max$200 a month (£152.04 / CA$286.54)YesEverything in Pro, plus significantly higher quotas
Teams$80 a month plus $40 a month per full developer seat (£60.82 plus £30.41 per seat / CA$114.62 plus CA$57.31 per seat)YesUp to 100 users, SAML and OIDC single sign-on, cost controls with three usage tiers, centralised billing, Slack channel support
EnterpriseCustomYesDedicated account management, SCIM and role-based access control, private networking, dedicated deployment option

Four pieces of fine print matter. First, the pound and Canadian dollar figures above use November 19, 2026 market rates of 1 USD = 0.7602 GBP and 1 USD = 1.4327 CAD, and Devin bills in USD, so they are approximations. Second, each paid plan comes with a usage allowance that refreshes daily and weekly, and going beyond it means buying extra usage at API pricing; the cost of a message varies with the model, the task size, and the reasoning required. Third, Max promises "significantly higher quotas" without publishing a number, so you cannot work out in advance how much Bot work $200 buys. Fourth, Cognition has published no per-Bot budget pricing, so a team running 20 Bots the way the game studio does cannot price that setup from the public page. SWE-2, Devin's latest model, is free to use on the Pro plan and above.

Devin Bots vs OpenAI Dots vs Anthropic Duet

The launch puts Cognition in direct competition with its own model suppliers. OpenAI's Dots works with Codex and Anthropic's Duet connects to Claude Code, and both shipped in the weeks before Bots. Our Claude Code explainer covers the tool Duet is built on.

Devin BotsOpenAI DotsAnthropic Duet
LaunchedNovember 17, 2026Late September 2026, as reportedLate October 2026, as reported
What it managesTeams of Devin coding agents, coding onlyTeams of agents on development tasks, inside CodexAgents connected to Claude Code, general-purpose
Model approachModel-agnostic, picks models per stage, plus Cognition's own SWE-2Built on OpenAI modelsBuilt on Claude models
Price for the manager layerNo standalone price; access from $200 a month (Max) or the Teams planNo separate price publishedNo separate price published
AvailabilityLive at devin.ai/bots; early access with dozens of customersShipping inside CodexShipping with Claude Code

Cognition's argument is focus. Dots and Duet are general-purpose agents that also handle code. Devin Bots are built around a coding agent that has been in production since 2024, and they coordinate Devin instances specifically. Wu's bet is that as models get smarter, the product built around them matters more, not less. The counterargument is that the frontier labs can absorb this capability into Codex and Claude Code, and Cognition's customers already pay those labs for the models. Both arguments are live, and the next year will settle which one holds. For a closer look at the other side of this race, our Mistral Vibe Desktop review covers another agent that runs several coding agents in parallel, from a lab that also builds its own models.

The honest catches

First, every launch number is Cognition's own. The 1,134 pull requests, the 70% lag cut, the 50-to-1,900 weekly jump, and the tenfold usage growth across early-access customers in two weeks all come from the company or its chosen customers. No independent test exists, and the tenfold growth is from a small early-access base of dozens of customers, not thousands of teams. 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 the bank cut tech vendor spend by 30% year over year as it brought work in-house. That is the existing Devin product, not Bots, but it is the closest thing to a real-world result in the whole story.

Second, a Bot that merges thousands of changes a week moves the review burden, it does not remove it. Somebody still has to read the code, set the rules for what auto-merges, and decide what the Bot is allowed to change on its own. For a small team, one engineer directing a Bot that directs a team of agents might be a real win. 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 and every author is an agent? Rules are only as good as the people who write them, and the failure mode is not a Bot that writes bad code, it is a team that sets auto-merge too loosely and finds the problem weeks later, in production, across a project the Bot has been quietly updating.

Third, you probably cannot try one today. Bots are in early access with dozens of customers, including Decagon, Fireworks AI, and Reducto, and Cognition has given no public date for wider access. The product page is live and sign-up is open, but early access is the reality for now, and the price that matters, what a Bot costs to run, is still unpublished beyond the plan it sits in.

Fourth, the technology behind the always-on behaviour is partly acquired, not built in-house. Fortune reports that Bots incorporate technology from The Interaction Company, the startup behind the Poke assistant, which Cognition acquired in July 2026. That is not a flaw, but it is worth knowing when judging how mature the persistent-agent layer is.

Who should care, and who should wait

If you already use Devin, this is your announcement. Bots are the upgrade path for the work Devin already does: longer projects, less babysitting. Watch for the early-access expansion and the per-Bot 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 work small teams usually cannot staff, and 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: 14,000 engineers and a 30% vendor-spend reduction is a real-world result for the existing product, so wait for a customer you can talk to before betting a roadmap on Bots. If you use Codex or Claude Code today, nothing changes yet. Dots and Duet already do the multi-agent version inside the tools you use, so Devin Bots is a reason to compare, not a reason to switch. If you want a smaller taste of Devin first, Cognition also ships a free Devin plugin that patches security flaws.

Common mistakes before you hand a Bot a goal

Cognition's own launch post is clear about what the human still has to do, and getting it wrong is the fastest way to waste $200 a month. Define what success means before you assign a goal: a Bot responsible for making CI faster needs to know which checks must remain and whether spending more on compute is acceptable. Build the feedback the Bot will judge itself by: tests, benchmarks, and user flows in the lab, and production behaviour and customer feedback in the wild. Set the auto-merge rules narrowly at first and widen them only after you have seen what the Bot merges on its own. Start with one Bot and one goal, not twenty, and read the Needs You inbox every day until you trust the Bot's judgment. And give the Bot a goal, not a backlog: "keep Android at parity with iOS" works because the Bot can decide what to build next, while a list of tickets just turns it back into an expensive ticket-taker.

Frequently asked questions

What is a Devin Bot?

A Devin Bot is a persistent agent from Cognition that owns an engineering goal and leads teams of Devin coding agents to deliver it over days or weeks. It keeps memory across sessions, works through Slack, iMessage, or the Devin app, and brings a human only the decisions that need judgment.

How much do Devin Bots cost?

There is no standalone Bots price. Access is included in Devin's Max plan at $200 a month (£152.04 / CA$286.54) and in the Teams plan at $80 a month plus $40 a month per full developer seat. The Pro plan at $20 a month does not include Bots. Extra usage beyond the plan allowance is billed at API pricing.

Can a Devin Bot merge code without a human approving it?

Only if a team allows it. Administrators set which changes require review and can limit auto-merge to low-risk edits. Code still passes the same tests used for human-written changes, and Bots run on separate cloud computers rather than on engineers' laptops.

How are Devin Bots different from OpenAI Dots and Anthropic Duet?

Dots and Duet are general-purpose always-on agents that also handle code, built into Codex and Claude Code. Devin Bots are coding-only, built around the Devin coding agent, and model-agnostic, picking different models for different stages of a project. All three launched within about seven weeks of each other.

When will Devin Bots be available to everyone?

Cognition has not given a date. Bots launched on November 17, 2026, in early access with dozens of customers, and the product page is live at devin.ai/bots with sign-up open.

Is there a free way to try Devin Bots?

No. The Free plan at $0 and the Pro plan at $20 a month do not include Bots. The cheapest route in is the Max plan at $200 a month, or the Teams plan for a group.

Devin Bots is the clearest version yet of the 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, and the model-agnostic routing are the marks of a product built for teams that actually ship. But the launch is a promise, not a result, and the two facts that decide whether this is a product you can buy, the price that matters, what a Bot costs to run, is still missing. Until they land, the honest move is to watch the early-access customers, keep the auto-merge rules tight, and judge Bots on the code they ship, not the pull requests they count. If you want the comparison in full, start with our Claude Code explainer and the Mistral Vibe Desktop review, then check Devin's pricing page yourself before you pay.

Sources

15 min read
39 views
edit_note Written by

Ratul Ahamed

Founder & Editor-in-Chief

Ratul Ahamed is the founder and editor of AI Core Updates, an independent site covering AI tools, models and pricing. He is a web developer based in Bangladesh and also runs All In One Tools, a free collection of browser-based utilities. Contact: [email protected]

Read all stories by Ratul Ahamed arrow_forward
verified_user Human-Reviewed Editorial

forum Comments (0)

Leave a Reply

chat_bubble_outline

No comments yet. Be the first to join the conversation!

Weekly AI Digest

Get our weekly AI tools digest

One email a week analyzing new AI tools, model updates, and practical workflows. Tested and summarized with zero fluff. Unsubscribe any time.

bolt
Hands-on AI Tool Reviews
Pricing math, limits, and alternatives.
verified_user
Strict Anti-Spam Policy
1-click unsubscribe at any time.
mark_email_unread

Subscribe to Digest

Sent weekly • No spam

person_outline
mail_outline

No spam ever. You can unsubscribe with a single click at any time.

cookie

Cookie Settings

We use cookies to enhance your browsing experience, serve personalized content, and analyze our traffic. By clicking "Accept All", you consent to our use of cookies. Learn more