Claude Code routines move agent jobs to the cloud with API and GitHub triggers

Anthropic’s April 14 X post looks like a product tweak on the surface, but the bigger shift is architectural. Claude Code routines move agent work off the user’s laptop and into Anthropic-managed cloud sessions that can keep running after the local machine is closed. That changes Claude Code from something you actively drive in a session into something that can sit behind recurring engineering work and fire when an external trigger appears.

‘Now in research preview: routines in Claude Code. Configure a routine once … and it can run on a schedule, from an API call, or in response to an event.’ Source tweet

The linked documentation gives the operational detail. Anthropic defines a routine as a saved Claude Code configuration: a prompt, one or more repositories, and a set of connectors, all packaged to run automatically on Anthropic-managed cloud infrastructure. The docs say routines support 3 trigger types—schedule, GitHub event, and API—and are available on Pro, Max, Team, and Enterprise plans with Claude Code on the web enabled. Anthropic also positions them for unattended work such as backlog maintenance, bespoke PR review, docs drift detection, and even cross-SDK porting. The reference page is here.

That matters because @claudeai has spent the last few months turning Claude Code into more than a chat-driven coding assistant. Routines add a durable workflow primitive. According to the docs, they run as full Claude Code cloud sessions with no approval prompts during execution, can use shell commands, and can call whichever connectors the user includes. The tradeoff is explicit: the routine acts through your linked accounts, and every run counts against your daily allowance. Convenience is rising, but so is the need to scope repositories, secrets, and branch permissions carefully.

The next question is whether engineering teams trust this enough to replace internal bots and hand-rolled CI glue. GitHub-triggered routines that open PRs, triage issues, or keep parallel SDKs in sync could be genuinely useful, but only if the permission model proves controllable in practice. If Anthropic gets that balance right, routines may end up being the feature that turns Claude Code from a tool people try into infrastructure teams keep running.