Claude Code Routines: Run Your AI Agent While You Sleep

On April 14, 2026, Anthropic shipped two things at once: a completely redesigned Claude Code desktop app and a new feature called Routines. The desktop redesign is a quality-of-life upgrade. The Routines feature is a different kind of shift — it's Anthropic's answer to the question 'what if your AI coding agent could work even when your computer is off?'

What Are Routines?

A Routine is a saved bundle: a prompt, one or more repositories, and any connectors you want (like GitHub or Slack). Once saved, a Routine can run on Anthropic's cloud infrastructure on a schedule or when you trigger it manually. Your laptop doesn't need to be on. You don't need to be at your desk.

The practical uses are immediately obvious: scheduled dependency checks, automated PR summaries, nightly test runs, weekly changelog generation, or any repeated task you'd normally fire off and wait for. Set it up once, and it runs on its own.

How to Create Your First Routine

Setting up a Routine in Claude Code:
1. Open the Claude Code desktop app (updated to latest version)
2. Click the new sidebar icon → Routines (currently in research preview)
3. Click 'New Routine'
4. Give it a name (e.g. 'Weekly dependency check')
5. Connect your repo via the GitHub connector
6. Write your prompt — be specific:

   'Check all npm dependencies in package.json for available updates.
   For each outdated package, note the current version, latest stable version,
   and whether it is a minor or major version jump. Flag any packages with
   known CVEs in the latest version. Output a summary in Markdown.'

7. Set a schedule (e.g. every Monday at 9am) or leave as manual trigger
8. Save — the Routine runs on Anthropic's cloud, not your machine

What the Desktop Redesign Adds

The new desktop app is more than a visual refresh. The headline addition is a sidebar that lets you run multiple Claude sessions side by side in a single window — no more switching between terminal tabs to manage parallel conversations. A built-in HTML and PDF preview panel means you can see rendered output without leaving the app. The diff viewer is faster and easier to read.

Version 2.1.110 (released April 15) added two commands worth knowing: /tui opens a full-screen terminal interface for distraction-free coding, and /focus switches to a flicker-free rendering mode for presentations or screen sharing. Ctrl+O now toggles between normal and verbose transcript output.

Prompts That Work Well as Routines

Not every task belongs in a Routine. The best candidates are tasks that are periodic (not one-off), clearly scoped (the AI knows when it's done), and output a report or artifact (rather than interactive feedback). Here are some that work well:

Routine ideas and prompts:

1. Weekly PR summary:
'Summarize all merged pull requests from the last 7 days in this repository.
For each PR, note the title, author, key files changed, and a one-sentence description of what it does. Group by area (frontend, backend, infrastructure). Output as Markdown.'

2. Daily test failure digest:
'Check the CI run logs from the last 24 hours. List any failing tests, the error message, and which files are involved. Flag any tests that have failed 3+ times in the past week.'

3. Monthly dead code scan:
'Identify functions and classes in /src that appear to be unused — not called anywhere in the codebase and not part of the public API. List each one with its file path and line number so a developer can review them.'

The Bigger Picture

Routines are Anthropic's clearest move yet toward autonomous agents that don't require you to be present. Combined with the desktop redesign's multi-session sidebar, Claude Code is shifting from a tool you talk to into infrastructure you configure. That's a meaningful difference — and it puts Claude Code in direct competition with GitHub Actions and other CI/CD workflows for a certain class of automated development tasks.