Overview
- Anthropic introduced routines for Claude Code on Tuesday, letting users save an automation that bundles a prompt, code repos, and connectors, then run it by schedule, API call, or GitHub webhook.
- Routines execute on Anthropic’s cloud without approval prompts during a run and can run shell commands, use repo-based skills, and call external connectors, with all actions tied to the initiating user’s identity.
- The feature is in research preview for Pro, Max, Team, and Enterprise users, with daily caps of 5, 15, and 25 runs respectively, and each run counts against token limits that can spike with large codebases.
- Anthropic also rolled out a redesigned desktop app that adds a multi-session sidebar, an integrated terminal, an in-app editor, a faster diff viewer, drag-and-drop panes, and a side chat that does not disrupt an active task.
- Developers and analysts frame routines as a shift from a chat assistant to an execution layer for tasks like CI checks, alert triage, and docs drift reviews, and early plugins show teams are already testing always-on workflows.