Anthropic launched Claude Code Routines into research preview on 14 April, giving developers a way to turn any Claude Code workflow into a scheduled or event-driven automation that runs entirely on Anthropic's managed infrastructure. A routine bundles a prompt, one or more repositories, and a chosen set of connectors into a reusable configuration. Once saved, it can fire on a schedule, in response to a webhook, or on demand via a POST to the API — and it keeps running whether or not the developer's laptop is open.
The three trigger mechanisms each open a different class of workflow. Scheduled routines handle recurring jobs on hourly, nightly, or weekly cadences — nightly backlog triage that labels and assigns new issues before morning standup, weekly documentation drift scans that open update PRs for APIs that have changed, or dawn-of-day dependency audits. Webhook routines, launching first with GitHub integration, subscribe to repository events so that a push, PR, or issue can kick off a review or generation task automatically. API routines accept a POST request and return a session URL, giving developers a clean integration point for alerting systems, deployment pipelines, and internal tools that need to invoke Claude Code on demand.
The feature is live now for Pro, Max, Team, and Enterprise subscribers with Claude Code on the web enabled, with daily run caps of 5, 15, and 25 respectively for Pro, Max, and Team/Enterprise tiers. Alongside routines, Anthropic also shipped a redesigned Claude Code desktop experience — multiple sessions side-by-side, a new session-management sidebar, an integrated terminal, file editing, HTML and PDF previews, a faster diff viewer, and a drag-and-drop customisable layout.
For context engineers, routines represent a meaningful shift in how Claude Code fits into real engineering workflows. Until now, every Claude Code session required a developer actively driving it — which made the tool brilliant for interactive work but awkward for the kind of repeatable, behind-the-scenes automation that CI pipelines and scheduled jobs have always handled. By hosting execution on Anthropic's own infrastructure and exposing three distinct trigger surfaces, routines bring Claude Code into the same operational tier as GitHub Actions or a managed cron runner — with the model's full agentic reasoning attached. The 5–25 runs-per-day caps keep the research preview conservative, but the architectural direction is clear: Claude Code is being engineered to run on its own.
