Your first scheduled task
A scheduled task runs by itself at fixed times – the simplest way into automation. Which tasks are suitable, how to set them up in ChatGPT and Claude and what to check in the first week.
Boaz LichtensteinFacts checked on 10 November 2026
Everything you have built in this academy so far starts with you: you open the chat and give the brief. A scheduled task turns that round. It starts at a fixed time, works without you and puts the result where you will find it.
What is suitable
Four conditions, and all four must hold:
- Regular. At least weekly – otherwise you learn nothing from the runs.
- Clear inputs. The sources are fixed and reachable.
- Checkable result. You can tell in five minutes whether it is right.
- Cheap mistake. A wrong run produces a wrong draft, not damage.
Good first candidates: the weekly report from order data, a Monday overview of appointments and open commitments, a summary of new entries in a source you follow, a reminder of deadlines from a list.
In four steps
1. Once in chat. Give the brief by hand and look at the result.
2. Sharpen. What was missing, what was too much? Repeat until you would pass the result on as it is. OpenAI’s own walkthrough recommends exactly this order.
3. Schedule. Tell the assistant when the task should run again – “every Monday at 7 am” – or create it in the overview of scheduled tasks.
4. Set a review slot. Put the first three runs in your calendar.
Where this works today
ChatGPT has a dedicated “Scheduled” page where you view, edit, pause and delete tasks. The limits depend on the plan:
Claude offers scheduled tasks in Cowork on all paid plans: hourly, daily, on weekdays, weekly or on demand. Every run opens its own session with the same connectors and skills as your normal work. New tasks on Pro, Max and Ultra have run in the cloud since 6 October 2026 – that is, even when your computer is off. Scheduled tasks in Cowork
The always-on agents from the previous path can schedule too. For getting started, the plain scheduled task is better: it does exactly one thing, and you see every run individually.
The first week
- Read every run. Not skim – recalculate one figure, check one statement against the source.
- Watch for what is missing. The most dangerous run is the one that looks good although a source could not be reached.
- Note every correction. Whatever you change by hand twice belongs in the brief.
What you should know
Scheduled tasks use quota even when nobody reads the result. Unattended tasks can be paused by the provider if you do not respond for a while. And a run only sees what is reachable at run time: a file that lives only on your switched-off laptop does not exist for a cloud run.
The next station takes on the brief itself – because a prompt that works in chat because you can ask follow-up questions is not enough for a run without supervision.