Anthropic shipped a feature this week that changes how software teams can think about AI-assisted development. Claude Code Routines, released April 14, 2026 in research preview, lets you set an AI coding workflow running once and have it execute on Anthropic’s cloud infrastructure automatically, whether that is on a nightly schedule, triggered by an HTTP API call, or fired by a GitHub event like a pull request opening.

The short version: you write a prompt, attach it to one or more code repositories, configure a trigger, and Claude runs the whole thing unattended. No open browser tab. No sitting at your laptop waiting for it to finish.

What Routines Actually Do

A routine packages three things together: a prompt, repositories for Claude to work in, and any MCP connectors you have set up (Slack, Linear, Google Drive, and so on). Once you save a routine, it becomes an autonomous task that runs on Anthropic’s servers under your account.

Three trigger types are available:

Scheduled runs the routine on a recurring cadence. You can pick hourly, daily, weekdays only, or weekly. For custom intervals, you can set a cron expression via the CLI with a minimum interval of one hour.

API gives each routine a dedicated HTTP endpoint. Send a POST request with a bearer token and Claude starts a new session immediately. You can pass additional context in the request body, for example an alert body from a monitoring system or a log file from a failed deploy.

GitHub fires the routine when specific events happen in a connected repository, such as a pull request opening, a release being published, or a PR being merged. You can filter on things like the PR author, labels applied, whether it is a draft, or whether it came from a fork.

A single routine can combine all three trigger types. A code review routine, for instance, might run nightly to catch any open PRs, also fire from your deploy script via the API after each production push, and react to every newly opened PR via the GitHub trigger.

The Use Cases Are More Useful Than They Sound

Anthropic’s documentation gives six example routines, and they read like a list of tasks that currently exist on an engineer’s plate and get done inconsistently:

• Nightly backlog triage: read issues opened since the last run, apply labels, assign owners, post a Slack summary before the team starts their day
• Alert triage: monitoring fires a webhook, Claude pulls the stack trace, finds the relevant recent commits, and opens a draft PR with a proposed fix
• Code review: Claude applies your team’s checklist to every new PR, leaves inline comments on security and style issues, and posts a summary so human reviewers can focus on design questions
• Deploy verification: after each production deploy, Claude runs smoke checks and posts a pass or fail before the deploy window closes
• Documentation drift: weekly scan of merged PRs flagging any docs that reference APIs that have changed
• Library porting: when a change merges in one SDK, Claude ports it to a parallel SDK in another language and opens a matching PR

None of these tasks require a human to do them manually every time. They require a human to set them up once and trust that they run.

What This Means for Business

For development teams, this is a meaningful step toward AI that earns its keep outside of interactive sessions. The value of AI tools has historically been measured in how much faster they make a developer while that developer is sitting at a keyboard. Routines extend that into time when no one is at the keyboard at all.

The practical implication for business leaders is that AI is no longer just augmenting individual effort. It is starting to fill in the spaces between people: the overnight reviews, the post-deploy checks, the weekly audits that almost always get deprioritized. Teams running Claude Code Routines can potentially ship with fewer regressions, tighter documentation, and faster PR turnarounds without adding headcount.

Routines are available on Claude Code’s Pro, Max, Team, and Enterprise plans. They are in research preview, which means the feature’s behavior, limits, and API surface may change. Daily run allowances apply per account: Pro users get five runs per day, Max gets fifteen, and Team and Enterprise users get twenty-five.

For data and engineering teams already using Claude Code, this is worth testing. The automation potential is real, and the setup cost is low. One well-crafted routine can replace a recurring manual task that costs an hour of a senior engineer’s attention every week.

If you want the playbook other teams are using with Claude and Codex right now, grab the free Working With Claude field guide. Download it here.