Cloud Parts

Devin

Cognition’s cloud coding agent: each session gets its own VM with a shell, editor and browser, takes a task from Slack, a ticket or the web app, and comes back with a pull request.

What it is
AI coding agents (cloud agents)
Type
Commercial product
Made by
Cognition
License
Proprietary
Facts checked
9 October 2026
Website

What it is

Devin runs in Cognition’s cloud, not on your laptop. Every session gets its own machine with a shell, an IDE and a browser, and your repository cloned in it, so it can install dependencies, start servers, run tests and click through the app it built before opening a pull request. You can watch the shell, take over in the embedded IDE, or leave it alone and review the PR later. Tasks arrive from Slack or Microsoft Teams threads, Linear or Jira tickets, the web app or the API, and Devin replies to review comments on its own PRs.

It learns about your code in several ways. It reads AGENTS.md before it starts coding (the first 16 KiB of each file go into its context automatically), uses an organisation-wide Knowledge base of instructions, and indexes repositories every couple of hours into a generated wiki. MCP servers, over stdio, SSE or HTTP, connect it to tools such as Sentry, Datadog or a database. Devin 2.0 (April 2025) added parallel sessions, each with its own cloud IDE, and an interactive plan you can edit before Devin works on its own.

Cognition also bought the Windsurf editor: it signed a definitive agreement to acquire Windsurf in July 2025, and Windsurf 2.0 (April 2026) lets you plan locally and send the work to Devin with one click. The Devin CLI, released the same month, is a local terminal agent whose /handoff command moves a session, with its context and git branch, to a cloud Devin.

Why teams pick it

  • Works asynchronously and in parallel: sessions keep running after you close your laptop, and many can run at once
  • Its own sandboxed machine per session, so it can’t damage your laptop and can test in a real browser
  • Fits a team’s existing tools: Slack and Teams, GitHub, GitLab and Bitbucket, Linear and Jira
  • Reads AGENTS.md and connects to MCP servers, so setup made for other agents carries over

Watch out for

  • Closed source and paid, and your code and secrets have to be available in Cognition’s cloud environment
  • Best on well-scoped tasks: Cognition’s rule of thumb is work a person could do in about three hours, with large tasks broken down
  • The machine has to be configured to match your development environment before results are reliable
  • Early benchmark claims need care: the launch figure of 13.86% on SWE-bench came from a random 25% of the tasks

How it fits with other technologies

See it in a real stack

No architecture uses Devin yet. Drop it into the playground and wire it up yourself.Open playground →
Where it came from2024–2026

Cognition introduced Devin on 12 March 2024, in a post by Scott Wu, as “the first AI software engineer”. Instead of completing code in an editor, Devin got a sandboxed computer with a shell, a code editor and a browser, and planned and executed tasks requiring thousands of decisions. On SWE-bench, real GitHub issues from projects like Django and scikit-learn, Cognition reported that Devin resolved 13.86% end to end, unassisted, against a previous best of 1.96%; it was evaluated on a random 25% subset. Cognition had raised a $21 million Series A led by Founders Fund, and Devin stayed in early access with a waitlist.

Devin became generally available on 10 December 2024 at $500 a month for engineering teams, with Slack as the main way to start sessions, an IDE extension and an API. Cognition’s advice was to give Devin tasks you know how to do yourself, tell it how to check its work and keep sessions under about three hours. Devin 2.0 followed in April 2025 with an agent-native cloud IDE, parallel sessions and a plan starting at $20. In July 2025 Cognition agreed to acquire Windsurf, then an IDE with $82 million of annual recurring revenue and more than 350 enterprise customers, and in 2026 it joined local and cloud work: Devin inside Windsurf 2.0 and a Devin CLI that hands sessions to the cloud.

Why it’s built this way

Give the agent a whole computer
A shell, editor and browser in a sandbox let Devin set up environments, reproduce bugs and test fixes end to end, the way a person would, instead of only proposing code.
A teammate you message, not a tool you watch
Slack was the primary interface at launch: you tag Devin on a thread and review its PR when it says it’s done, which suits work that doesn’t need you at the keyboard.
Plan together, then let it run
Scoping a task can take as long as doing it, so Devin 2.0 researches the codebase and proposes a plan you can edit before it works on its own.
Start local, hand off to the cloud
Cognition argues a local agent makes you faster while a cloud agent works while you’re away, so in 2026 it connected the two with Windsurf and the Devin CLI.

Timeline

  1. 2024

    Introduced as “the first AI software engineer” (March)

  2. 2024

    Generally available at $500 a month for teams (December)

  3. 2025

    Devin 2.0: cloud IDE, parallel sessions and a $20 plan (April)

  4. 2025

    Cognition signs an agreement to acquire Windsurf (July)

  5. 2026

    Devin built into Windsurf 2.0; Devin CLI hands local sessions to the cloud (April)

Sources10 · facts checked 9 October 2026

Improve this page on GitHub