Anthropic Unveils Claude Code Routines for Autonomous Developer Workflow Automation

Anthropic has introduced Routines for Claude Code, a feature currently in research preview designed to automate software engineering workflows. These autonomous agent sessions run on Anthropic managed cloud infrastructure, allowing tasks to execute even when a developer computer is turned off. By leveraging natural language prompts and repository access, Routines aim to eliminate the need for manual intervention in repeatable maintenance tasks.

The system supports three distinct trigger types to initiate sessions. Scheduled triggers run on a recurring cadence, such as nightly or weekly, while API triggers allow for on-demand execution via HTTP requests. Additionally, GitHub event triggers can automatically start a session in response to pull requests or new releases, providing a seamless bridge between code changes and AI actions.

Practical applications for Routines include triaging issue backlogs, generating documentation updates, and performing bespoke code reviews. The tool can also be integrated into deployment pipelines to run smoke tests or scan error logs for regressions. This functionality is supported by Model Context Protocol connectors, which give Claude access to external services like Slack and Linear. Availability is limited to Pro, Max, Team, and Enterprise subscribers who have Claude Code on the web enabled. Users can manage these automated tasks through a centralized dashboard at claude.ai/code/routines or via the command line interface. Each routine operates as an independent session, with actions attributed directly to the user identity. Security remains a priority, with Routines restricted to pushing code to specific branches by default to prevent accidental modifications to protected codebases. Usage limits are also in place, including daily caps on the number of runs and hourly limits for GitHub webhooks. Organizations requiring more volume can enable metered overage through their billing settings.