Anthropic Ships Claude Code Routines — Your Agent Now Runs When Your Laptop Is Closed
Anthropic launched Claude Code Routines in research preview today — saved Claude Code configurations that run on schedules, respond to GitHub events, or trigger via API, executing on Anthropic-managed cloud infrastructure without requiring an open session.
Original sourceAnthropic today released Claude Code Routines in research preview, a feature that fundamentally changes how Claude Code can operate. Instead of requiring an active user session, Routines are saved configurations — a prompt, one or more repositories, and a set of MCP connectors — that execute autonomously on Anthropic's cloud infrastructure on a recurring schedule, in response to GitHub events, or triggered by an authenticated API call.
The three trigger types cover the core automation scenarios teams actually need. Schedule triggers run on preset cadences (hourly, daily, weekdays, weekly, or custom cron expressions with a minimum one-hour interval). GitHub triggers react to repository events — pull requests opened, pushes, issues, workflow runs, or any of 15 other event categories — with optional filters for PR author, labels, draft status, and more. API triggers give each routine a dedicated HTTP endpoint: POST with a bearer token and an optional 'text' field to inject context (like an alert body or stack trace), and Claude starts a session immediately.
The practical use cases Anthropic demos are real developer pain: nightly backlog maintenance that reads issues and applies labels, alert triage where a monitoring tool fires the routine and passes a stack trace for Claude to correlate with recent commits, bespoke code review that applies team-specific checklists on every PR open, and library porting between SDKs in different languages triggered by merged PRs.
Security boundaries are defined but narrow. By default, Claude can only push to branches prefixed with 'claude/' — preventing routines from accidentally modifying protected branches. Connectors give access to external services like Slack, Linear, and Google Drive, but scope should be set to what the routine actually needs. Commits and PRs created by routines appear under the user's GitHub identity, which carries real accountability implications. Routines are individual to the claude.ai account that created them and cannot be shared with teammates.
The feature is available on Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled, and draws down subscription usage alongside interactive sessions with a separate daily run cap. The API surface ships under an explicit experimental beta header with documented breaking-change policies — Anthropic will maintain the two most recent previous header versions during migrations. This is early infrastructure, not a stable API, and Anthropic is signaling as much clearly.
Panel Takes
“GitHub triggers that auto-run PR reviews, schedule triggers for nightly backlog grooming, and API triggers that plug into your existing alerting pipeline — this is the missing piece for teams that wanted Claude Code to work like a teammate that's always on. The API trigger with a context-injecting 'text' field is elegant: your PagerDuty alert fires the routine and passes the stack trace, and Claude has everything it needs. The 'claude/' branch prefix default is sensible guard rails.”
“Research preview with explicit warnings about breaking changes, daily run caps, per-account limits, and 'runs rejected until the window resets' language is not production infrastructure. The fact that routine commits and PRs appear as your GitHub user means when the routine confidently merges something wrong, that's your name on it. Autonomous agents writing to protected branches, even with the default 'claude/' prefix guardrail, is a high-trust scenario that needs more than a beta header in the API to earn that trust.”
“The architecture here — saved configurations with multiple trigger types, cloud execution so 'they keep working when your laptop is closed,' and connector access to external services — is the scaffolding for persistent AI teammates. This is not 'Claude does tasks when you ask.' This is 'Claude operates continuously on your infrastructure.' Routines are the first piece of Anthropic shipping Claude as a team member rather than a tool.”