Sulat.com / Writing on AIWednesday, November 18, 2026
Agent Engineering/18 November 2026

Cognition launches Devin Bots, agents that own an outcome and run teams of Devin sessions

Cognition says a Devin Bot takes an outcome, starts a Devin session for each piece and reports back in Slack, iMessage or the web app. Beta testers merged 4.5 times more PRs per message, the company says.

A 3D Devin logo at the centre of a dark control room, with light trails running to small robots at workstations showing green checks

At a glance

  • Cognition says a Devin Bot takes an outcome, starts a Devin session on its own cloud machine for each piece of work, and keeps going until the pull requests are ready to merge.
  • Cognition says beta testers merge 4.5 times more pull requests per message on average, and that individual engineers went from 50 PRs a week to nearly 2,000.
  • You can reach a Bot from the Devin web app, its own Slack app or iMessage, and Cognition's product page shows QA, Developer and PM Bots handing work to each other.
  • Cognition's Bot marketplace lists 16 templates, including Software Factory, QA Engineer and Chief of Staff, and a new Bot starts with the template's prompt and no connections, secrets or memory.
  • A shared Bot works with its own access, so everyone you add can use its sign-ins, secrets and sessions, and a Bot's usage bills to its creator.

Cognition has launched Devin Bots, agents that take an outcome such as “own checkout”, start a Devin session for each piece of work and keep going until the pull requests are ready to merge. Cognition announced them on X on 17 November 2026, saying beta testers merge 4.5 times more pull requests per message and that individual engineers went from 50 PRs a week to nearly 2,000.

Cognition’s post says: “Meet Devin Bots. From writing firmware to merging thousands of PRs a week, a Devin Bot leads a team of agents to get the job done.” A pull request, or PR, is a proposed code change that a person reviews before it joins the codebase. The 4.5x figure is an average across beta testers, according to Cognition.

What a Devin Bot does

According to Cognition’s docs, a Bot has one chat for everything it owns. You tell it the outcome you want there, and its reports come back there. The docs’ example is “You own the Android app now. Keep it at feature parity with iOS.” The Bot plans the work, starts a Devin session for each piece, follows up on those sessions and comes back to you when something needs you.

Each session runs on its own cloud machine, the same as a session you start yourself, the docs say. Before, one request started one session that opened one pull request. Now one request goes to a Bot, which connects tools such as Slack, GitHub and Linear and keeps as many sessions running as the work takes. Cognition’s product page shows a Bot given “Move checkout to the new payments API” splitting it into updating the API client, migrating webhooks and rewriting checkout tests, each on its own machine.

Before and after diagram: one request starts one session, while one outcome goes to a Bot that runs many sessions
One goal goes to the Bot, and the Bot starts as many sessions as the work needs.
PartWhat it does
ChatOne chat for everything the Bot owns; its reports come back there.
SessionsEach piece of work runs in its own child session with its own machine.
AutomationsThe Bot runs on a schedule or when something happens in a connected tool.
Needs youApprovals, decisions and pull requests to review collect in one list.
MemoryWhat the Bot learns is saved and used in later sessions.
AccessThe Bot has its own connections and secrets, separate from the people who message it.
The six parts of a Bot, from Cognition’s docs.

Where you talk to one

You can reach a Bot from the Devin web app, Slack or iMessage, the docs say. In Slack, each Bot can have its own Slack app and handle, so your team mentions it in a channel or messages it directly and it replies in the thread. If your workspace requires admin approval for apps, a Slack admin has to install the Bot’s app.

iMessage works through a one-time code. You open the iMessage page of the Bot’s settings, send the text it prepares, and that links your number to the Bot. Texts go to the same chat as your messages in Devin. iMessage isn’t available on Enterprise plans. Cognition’s product page adds that you can call a Bot for an update or bring one into a group chat.

Fortune reports that a Devin Bot joins a team’s Slack as its own member, remembers context across weeks and contacts people when it’s blocked or needs a decision. Fortune’s headline frames Bots as a challenge to Anthropic’s Duet and OpenAI’s Dots, and the paper reports that the product incorporates technology from The Interaction Company, the startup behind the Poke assistant, which Cognition acquired in July.

Cognition announced that deal on 23 July. Co-founder Scott Wu wrote then that the Interaction team had built an agent that’s “proactive, it knows you, and it’s fun to talk to”, and that “that’s exactly how working with Devin should feel”. Interaction co-founder Marvin von Hagen told TechCrunch: “I think there’s a lot of value in having Poke orchestrate different Devin sessions.”

Bots that hand work to each other

Cognition’s page describes a team that coordinates itself. A QA Bot finds that discount codes aren’t applied at checkout and asks the Developer Bot to take a look. The Developer Bot fixes it and opens a pull request, the QA Bot retests on staging, and a PM Bot tells the channel the launch stays on track for Thursday. Cognition’s own summary: “Your QA Bot finds a bug and brings in your Developer Bot to fix it, then tests the result.”

Four-step chain: QA Bot finds a bug, Developer Bot fixes it, QA Bot retests it, PM Bot reports back
QA and Developer Bots pass work between them while a PM Bot reports in Slack.

Handoffs run on grants. Giving Bot A access to Bot B lets A read B, including its conversation and sessions, and message it, the docs say. A can’t change B’s settings. The Chief of Staff template, for example, “watches your channels, posts the updates and hands work to the rest of the team”, then follows that work until it’s done.

Bots also set their own triggers. Cognition’s page lists watching a Slack channel for customer issues, verifying CI health every night at 2am and monitoring Linear for blockers. CI, short for continuous integration, is the set of automated checks that run on code before it merges.

Sixteen templates to start from

Cognition’s Bot marketplace lists 16 templates, and you can browse them without signing in. A new Bot built from a template gets the template’s prompt and nothing else. It’s private to you and has no connections, secrets or memory until you add them, per the templates page.

TemplateWhat it does
Software FactoryFinds bugs, polish, performance and refactoring work in a product area and opens one small pull request for each, with proof it works.
Product EngineerTakes a ticket or request all the way to a tested pull request and sees it through review.
QA EngineerTests each pull request in a real browser and posts the video.
Chief of StaffWatches your channels, posts updates and hands work to the rest of the team.
Latency OptimizerFinds your slowest endpoints, jobs and queries, and once a week tries a fix for the worst one.
Code ReaperFinds code nothing uses and removes it, one safe pull request at a time.
Migration EngineerOwns one big migration, shipped as small, tracked pull requests.
iOS EngineerBuilds SwiftUI features and ships them to TestFlight.
Eight of Cognition’s 16 Bot templates, shortened from the company’s descriptions.

You can also share your own Bot’s instructions as a template. On Teams and Enterprise it appears in your organisation’s library, and on Pro, Max and Teams you can copy a public link. The docs warn that a public link can’t be edited or revoked, and that Cognition reviews only its own templates.

What early users report

Jenny Xu, founder of the mobile game studio Talofa Games, posted on 17 November: “I used to have 50 PRs a week, now I have 1,900.” She says she built a bot called SuperJenny three weeks ago that works with about 20 AI teammates to ship her game Monster Walk every week, covering builds, QA, store reviews, release notes and analytics. She says the setup helped her grow the studio’s revenue sixfold.

Xu named five of the roles. SuperJenny delegates, SuperPM publishes a ranked daily fix list, SuperRelease builds weekly with a version ready to submit on Fridays, SuperQA runs real-device tests with video proof, and SuperSupport drafts every store review reply for her approval. She checks in four times a day from her phone. She released the setup free as SuperJennyStack, which is in Cognition’s template library.

Three stat cards: 4.5x more PRs merged per message, 50 to 1,900 PRs a week, 52 to 879 merged PRs a week after 19 days
Each figure comes from the person or company that posted it.

The jump isn’t new for Talofa. On 26 October the studio wrote that 19 days after it put Devin on Monster Walk, merged PRs per week went from 52 to 879 and 48 automations ran parts of the studio. Talofa is also running Devin for Gaming with Cognition, a three-week programme for game studios with a demo day on 19 November.

More early users on X

Early users are posting their setups. Avinash, Cognition’s regional community lead for MENA, borrowed an idea from Xu: a Chief of Staff bot that rewards the other bots every day. He says the system lets the bots self-improve, and suggests anyone new to Bots try it. Mark Fenner, an AI engineer and Devin ambassador, runs four Bots, led by a Chief Agent called Frankie that coordinates the rest and reports back to him. He says his Software Factory Bot, told to do a “Polish push”, scouted the codebase, split the issues into separate Devin sessions and returned fixes and pull requests. One of them fixed keyboard focus escaping behind a checkout window, and Devin Review flagged that the tests hadn’t covered the real Stripe fields.

Chris Wozniczek, who lists himself as a Cognition ambassador, says he mostly uses Bots in Slack to run his web studio, handle cold outreach and build previews for customers. Nick White, who founded Helionova and is also a Cognition ambassador, says he’s tested Bots for a few days and that the focus is on building with your own swarm of professional developers. Ariel Cárdenas posted in Spanish that he connected Devin to his home lab. From his phone he tells it which film he wants, one bot finds it, downloads it to Jellyfin and syncs the subtitles, and it’s ready when he sits down at the TV.

Access, security and who pays

A Bot works with its own access, not the access of whoever messages it, according to the access and security page. A new Bot is private to its creator. Share it, and everyone you add can use every sign-in, secret and session the Bot holds, by asking it. Access comes in three separate levels, Can use, Can configure and Can share, and none includes another.

A private Bot takes its creator’s sign-ins, plugins and secrets without a prompt. A shared Bot shows an approval card to the person it’s answering, who approves giving their own. Two rules tie each grant to the person who gave it: a grant never exceeds the giver, and a sign-in lasts only while the giver still has it.

Side by side: a private Devin Bot uses its creator's sign-ins without asking, a shared Bot shows each person an approval card
Sharing a Bot means each new piece of access needs someone’s approval.

Cognition’s product page shows SOC 2, ISO 27001 and FedRAMP badges. It lists comprehensive audit logs, zero data retention enforced with model providers, a separate VM for every session, role-based access control, SSO, SCIM, OIDC and private networking.

A Bot’s usage is its chat plus every session it starts, and its creator pays, per the usage limits page.

PlanWho paysCap
MaxThe creator’s included quota, then on-demand creditsNo separate limit for a Bot
TeamsThe creator’s included quota, then the team’s on-demand creditsOn-demand cap in dollars, set by the creator or admins; an admin can give the Bot its own limit
EnterpriseCounts toward the creator’s own usage limitACU or dollar limit; the Bot stops at it when budgets and user limits are enabled
Who pays for a Bot’s usage, from Cognition’s docs.

How to start one

  1. In the Devin sidebar’s Bots section, click Create new Bot.
  2. Pick a template or Start from scratch, then name the Bot.
  3. Tell it in the chat what it owns, such as a repository, project or channel.
  4. Set up its connections, and give it its own Slack app if your team will message it.
  5. Check Needs you for approvals, decisions and pull requests.

A Bot’s work counts toward your plan’s usage like any session. Cognition’s last launch before Bots was Devin Apps, private web apps hosted on Devin’s cloud computer, on 13 November.

Disclosure: I’m an OpenAI, Firecrawl and Devin ambassador.


Discover more from Sulat AI & Tech News

Subscribe to get the latest posts sent to your email.