Devin review

Devin is Cognition's autonomous AI software engineer. It writes, runs and tests code in its own virtual machine with a shell, IDE and browser, and prepares pull requests for review ^1,7. Cognition aims it at tickets, migrations, bug fixes and internal tools, and says a task a developer could finish in about three hours is a good fit ⁷. The Devin brand now also covers Devin CLI and Devin Desktop, the editor formerly sold as Windsurf ²⁹.

Cognition bought Windsurf in July 2025 ²⁸. In September 2026 it raised $2 billion at a $48 billion valuation, with annualized revenue near $900 million, up from $492 million in May ^26,27. TechCrunch lists Mercedes-Benz, NASA, Goldman Sachs and Citi among its major enterprise customers ²⁷.

You connect GitHub, GitLab, Bitbucket or Azure DevOps and configure an environment: a Linux VM snapshot with your repos, tools, dependencies and secrets that every session boots from ⁸. Windows and macOS snapshots are optional ⁸. Sessions start from the web app, Slack, Microsoft Teams, Linear or Jira, the API or the CLI, and you can take over in Devin's IDE at any point ⁷. MCP servers, installed from a marketplace or added as stdio, SSE or HTTP servers, connect Devin to tools such as Sentry, Datadog and Linear ⁹.

Locally, Devin CLI runs in modes from Normal, where every write and command needs approval, to Autonomous, where commands run inside an OS-level sandbox ^11,12. Allow, ask and deny rules cover reads, writes, commands, web fetches and MCP tools, and organization rules cannot be overridden ¹¹. The /handoff command moves a local session to the cloud with its branch and uncommitted changes, or pulls a cloud PR branch back to your machine ¹⁰. Devin Review checks pull requests on GitHub, GitLab and Azure DevOps ¹⁴.

In April 2026 Cognition moved self-serve plans from ACUs to token quotas, with overage billed in dollars ⁴. Free costs nothing, Pro is $20 a month with Devin Cloud and up to 10 concurrent sessions, and Max is $200 a month for a much larger weekly quota with no daily cap ^2,3. Teams has an $80 monthly minimum: full seats cost $40 and carry a Pro-sized quota, while free flex seats draw on a shared pool of on-demand credits ³. Quotas reset daily and weekly and are measured in tokens, so how far they go depends on the model and context you use ⁵. Past the quota you pay API list prices for the model, and purchased credits roll over ^3,5. Enterprise is billed in ACUs at the rate in your order form ⁶. Ask Devin's Deep Mode, Devin Review and premium DeepWiki generation are now usage-billed, though Devin Review stays free for open-source repos ⁴.

Cognition has held SOC 2 Type II since September 2024 and lists ISO/IEC 27001:2022 in its trust center, where reports sit behind an NDA ^19,21. Data is encrypted in transit and at rest and kept for the length of the customer relationship ¹⁹. Enterprise data is never used for training without written consent ¹⁹. On self-serve plans Cognition may use data for training by default; a paid workspace can opt out, which also enables zero data retention with model providers, and on Teams only an admin can do it ¹⁹. Another Cognition page says it does not train on customer code by default, so get the terms in writing ²⁰. Enterprises can run Devin in a single-tenant VPC reached over AWS PrivateLink or IPSec ²³. Devin Cloud is in process for FedRAMP High, IL4 and ITAR, which Devin CLI and Devin Desktop already hold ²².

Independent proof comes mainly from the press. Goldman Sachs CIO Marco Argenti said the bank would deploy hundreds of Devin instances, possibly thousands, under human supervision ²⁵. TechCrunch names Mercedes-Benz, NASA, Goldman Sachs, Citi and Santander as customers ^26,27. The same coverage notes that researchers found earlier versions struggled with complex coding work ²⁵. Cognition publishes customer stories from Mercedes-Benz, AngelList, Litera, Nubank and others, several with hard numbers ²⁴.