You don't have to be present for Claude to work. Schedule a task to run on a cadence ("every weekday at 8am, summarize yesterday's support tickets") or have it repeat on a loop until something is done. In Claude Desktop you set this up by asking in the chat, in plain language, just like any other request — no cron syntax. It runs, does the job, and leaves you the result — like a coworker who handles the standing chores overnight. (The terminal's /schedule and /loop commands are the Power-Track way to do the same thing.)
There are two shapes to this. A schedule runs a job on a cadence — “every weekday at 8am, summarize yesterday’s support tickets” — like a standing appointment Claude keeps without you. A loop runs the same task over and over until a condition is met — “keep checking the deploy every five minutes and tell me the moment it goes green.” One is for chores that recur on the clock; the other is for waiting on something to happen.
Setting one up is no harder than any other request. In Claude Desktop, you ask in
the chat in plain language — “every Monday at 9am, pull last week’s numbers and draft
the team update” — and Claude handles the timing; there’s no cron syntax to learn. (If
you live in the terminal, the Power-Track /schedule and /loop commands do the
same job.)
The shift here is that you no longer have to be present for the work. The job runs, does the thing, and leaves you the result — so the standing reports and morning summaries that used to eat the first hour of your day are simply finished when you arrive.
Two habits make this trustworthy. Save the output somewhere you’ll see it — a file, a doc, a message — so finished work is actually waiting, not lost in a log. And start with a task you already trust Claude to do by hand: prove it on the manual version first, then let it run unattended. This is the time-shifted cousin of a hook — a hook fires on an event, a scheduled agent fires on the clock.
why it helps Wake up to work that's already finished.
examples- On Desktop, just ask in plain language ("every Monday at 9am, …") — no cron syntax to learn.
- Schedule the standing chores — daily summaries, weekly reports — and reclaim that time.
- Have the job save its output somewhere you'll see it, so finished work is waiting for you.
- Start with a task you already trust Claude to do by hand before you let it run unattended.
Questions people ask
- What's the difference between a schedule and a loop?
- A schedule runs a job on a cadence — "every weekday at 8am, summarize yesterday's tickets" — like a standing appointment Claude keeps. A loop runs the same task over and over until a condition is met — "keep checking the deploy until it goes green."
- Do I need to be present for a scheduled agent to run?
- No — that's the point. The job runs on its own, does the work, and leaves you the result, so the standing reports and morning summaries are simply finished when you arrive.
- Do I need to know cron or the terminal to schedule a job?
- No. In Claude Desktop you set up a schedule by asking in the chat in plain language — "every Monday at 9am, draft the team update" — with no cron syntax. The terminal's `/schedule` and `/loop` commands are the Power-Track way to do the same thing.
- How is a scheduled agent different from a hook?
- A hook fires on an *event* (before a command, after an edit); a scheduled agent fires on the *clock* (every morning, every Monday). A scheduled agent is the time-shifted cousin of a hook.
- How do I make sure I see what a scheduled agent produced?
- Have the job save its output somewhere you'll actually see it — a file, a doc, a message — so finished work is waiting rather than lost in a log. And start with a task you already trust Claude to do by hand.