Devin
Cognition's hosted autonomous AI software engineer that takes engineering tasks end-to-end inside its own Workspace — shell, IDE, and browser — charged in Agent Compute Units (ACUs).
Description
Devin is a proprietary cloud agent positioned as an autonomous AI software engineer. Each session runs on a virtual machine where Devin can write, run, and test code, with a Workspace UI exposing three primary developer tools: a shell, an embedded IDE, and a browser. Cognition meters usage in Agent Compute Units (ACUs) covering planning, context gathering, task execution, browser actions, and code execution. Devin pulls cross-session context from a Knowledge store auto-populated from repo READMEs and from .rules / AGENTS.md / CLAUDE.md style files; Knowledge is retrieved per session by a Trigger.
Solution
Cloud-hosted single-agent session inside a Devin VM. Devin opens a Workspace with shell, IDE, and browser, plans the task, gathers context (including org Knowledge), and iterates: write code, run, test, browse docs, repeat. The session can sleep and resume when messaged. Usage accumulates in ACUs per action class (planning, context, execution, browser, code execution).
Primary use cases
- autonomous end-to-end engineering tasks in a hosted VM
- shipping pull requests via Slack/Linear/GitHub handoff
- incident-response and CI-failure triage as agent tasks
- long-running multi-hour tasks scoped at ~3-hour granularity
Open the full interactive page →
Diagram, neighbourhood map, code examples, related patterns and full provenance.