Anthropic adds 'routines' to Claude Code for scheduled agent tasks

Anthropic is adding a new “routines” feature to Claude Code, allowing developers to schedule and automate agent-driven coding tasks to run without direct interaction.

A routine, essentially, is a saved task — defined by a prompt, connected repositories, and available tools — that can be triggered on a schedule, via an API call, or in response to events such as GitHub webhooks. Once configured, it runs in Anthropic’s cloud environment, carrying out work without requiring an active session.

For example, a routine could be set to run every night at 2am, pulling in a bug report, attempting a fix, and opening a draft pull request, or triggered by a GitHub event to review changes and post a summary to a team channel.

The new routines dashboard in Claude Code looks a little something like this: define the task, choose a trigger, and attach the systems the agent needs, with execution happening in the background.

Zoom

Routines bring structure to automation

Claude Code already serves as an agent – it writes code, runs commands, and interacts with repositories. But these capabilities have largely been tied to an active session, where a developer initiates the work and follows its progress.

Even so, developers have been using it to automate parts of the software development cycle. Doing that has meant setting up and maintaining their own systems — from scheduled tasks such as cron jobs, which run commands at fixed times, to the surrounding infrastructure needed to keep everything running.

Routines move that setup into Claude Code itself. A task can be defined once and then executed whenever the conditions are met, whether that is a scheduled run or a repository event. Code review, dependency updates, and issue triage can be handled in this way without manual prompting each time.

Managed execution and orchestration

Routines place the execution layer inside Anthropic’s infrastructure, with scheduling, orchestration, and state handled as part of the service. That removes the need to build and maintain external systems to keep tasks running over time.

Once defined, a routine can continue to operate in the background, triggered by a schedule or an event rather than a manual prompt. That moves Claude Code closer to the kind of long-running, production systems Anthropic has clearly been embracing of late, where agents are configured to keep working over time rather than being restarted for each task.

Claude Code still supports interactive use, of course, but it can also run predefined tasks without direct input, handling repeat work as part of an ongoing system.

How to access routines in Claude Code today

Routines are available to Claude Code users on Pro, Max, Team, and Enterprise plans, provided Claude Code is enabled on the web. They can be set up through the web interface at claude.ai/code or via the CLI using the /schedule command.

Usage is counted in the same way as standard Claude Code sessions, with additional daily limits depending on the plan. Pro users can run up to five routines per day, Max users up to 15, and Team and Enterprise users up to 25.

Additional runs are possible beyond those limits through paid usage.