Anthropic just detailed 'Routines' for Claude Code, packaging configurations—prompts, repos, and connectors—into automated workflows running on Anthropic-managed cloud infrastructure. This means development tasks persist even when a developer’s laptop is offline. Routines offer a powerful way to operationalize AI agents beyond interactive sessions, moving into persistent background work.

Developers can hook routines up using three trigger types: scheduled cadence, direct API calls via an HTTP endpoint, or GitHub events like new pull requests or pushes. A single routine can combine these, such as a PR review process that runs nightly but also fires on demand. These features are available across Pro, Max, Team, and Enterprise plans.

Practical use cases demonstrate immediate value, from weekly backlog grooming posting summaries to Slack, to alert triage systems that automatically generate draft PRs with proposed fixes. Another example shows bespoke code review routines applying checklists inline upon PR opening. This moves Claude Code from a coding assistant to a true background automation engine.

Management surfaces exist across the web interface and CLI, where commands like `/schedule list` help maintain existing configurations. Since routines run as full cloud sessions, they execute shell commands and utilize repository skills autonomously, making them useful for cross-language library porting or verifying deployments.