Claude Code routines promise mildly clever cron jobs

What Anthropic shipped

Anthropic has introduced "routines," a cloud service that runs Claude Code automations on Anthropic-managed infrastructure. A routine, the company says, is a saved Claude Code configuration — a prompt, one or more repositories, and a set of connectors — packaged once and run automatically. It has been reported that Anthropic’s infrastructure “hasn't been all that reliable lately,” so the pitch is simple: let the company keep the job running even when your laptop is closed.

Not cron, not an agent — somewhere in between

Routines sit in an awkward, useful middle ground. Think cron jobs and GitHub Actions, but with an AI in the loop: scheduled or trigger-driven prompts that can inspect CI/CD output, triage alerts, or take different actions depending on what the model finds. They're not ongoing, stateful agents; they're short-lived, trigger-executed runs — dynamic cron jobs with a taste for context. Convenience? Sure. Control? Maybe less so. And yes, Anthropic apparently wants you to use its UI: it has been reported that the company prefers customers interact through its app rather than third-party plugins — allegedly OpenCode has already been cut off from subsidized subscription usage.

Limits, a redesigned app, and a pragmatic caveat

Routines are gated by plan: Pro users get five routine runs per day, Max 15, and Team/Enterprise 25, with metered overage available if you enable it. Anthropic also refreshed its Claude Code desktop app (still Electron-based) with an integrated terminal, file editor, faster diffs, and expanded preview — the whole “ship without bouncing to your editor” move. Nice. Handy. And a little like giving someone a Swiss Army knife that eats tokens: multitasking across repos burns through model context fast, and that translates to real cost. So ask yourself: is the extra polish worth the throttle? For many teams, the answer will be yes — until the bill arrives.

Sources: The Register