Until now Claude only did something when you typed a request. In this tutorial you'll set up your first loop and your first cloud routine yourself — and learn which to reach for. Follow along in Claude Code; about ten minutes.
The whole trick is one question: what sets it off?
Repeats one task on an interval — every 10 minutes, every hour — while it runs. You start it; it keeps going until you stop. Runs in your open session.
Runs at a set time on its own — every morning at 8 — up in Anthropic's cloud, even when your laptop is closed.
Reacts the moment something happens inside your work — a task finishes, a file changes. Set up once (a hook), it fires every time. A prompt is a request; a hook is a guarantee.
The more you want it to run on its own, the less your own laptop needs to be on. A loop leans on your open laptop; a cloud routine runs without you.
Open Claude Code and type /loop 20m check for new replies and tell me. You'll see Claude do the task, then a line like "next run in 20:00" — it just repeated itself, on its own. Press Esc to stop. Leave the time out (/loop check for…) and Claude picks its own pace.
Type /schedule daily at 8am, [your task] — or open claude.ai/code/routines and click New routine. Pick a time, write the task, hit Create. Now it runs every day on Anthropic's servers — even with your laptop shut. Each run shows up under your routines.
A trigger fires on a built-in event — like Claude finishing a task. You set it once as a hook in settings.json. For an outside event (a new listing on a website) there's no ready-made trigger — you check with a loop or routine. We'll go deeper another week.
Pick one task you do by hand today. Decide: does it need to repeat while you watch (loop), or run on a schedule (routine)? Then fill in the prompt and run it.
You've run one loop and scheduled one routine for your own business — and you know which to reach for next time.