Will a scheduled AI task run if my computer is asleep?

A scheduled task is an alarm clock that does a job when it rings. The question that matters is where the clock lives. If it lives in the company’s cloud, your laptop can be shut in a drawer. If it lives on your laptop, the laptop has to be awake when it rings.

“Scheduled” sounds the same in every app. It is not.

The short answer

Where you set it up	Runs on	Computer must be awake?	Can it read files on your computer?
ChatGPT scheduled task	ChatGPT, online	No	No
Claude Code cloud task	Anthropic’s cloud	No	No, it works from a fresh copy of your project online
Claude Desktop scheduled task	Your computer	Yes, and the app must be open	Yes
/loop in Claude Code	Your computer	Yes, and that session must stay open	Yes

Checked against OpenAI’s and Anthropic’s help pages on 21 October 2026. These products change often; the help pages are the final word.

What a sleeping laptop actually does

Claude Desktop tasks skip the run. When the computer wakes, it runs once to catch up on the most recent missed time and drops anything older. So a 9am task can turn up at 11pm if the laptop slept all day. There is a “Keep computer awake” setting, but closing the lid still puts it to sleep.

/loop is for watching something while you work (“check the build every five minutes”). It only runs while that Claude Code session is open and idle, it does not catch up on missed runs, and recurring loops delete themselves after seven days.

Cloud tasks, in ChatGPT or Claude, do not care. They run whether your computer is on, off or in a bag.

How to choose

• It only needs the internet or connected apps (a morning news summary, a weekly email digest): use a cloud task. This is what your first automation sets up.
• It needs a file or program on your computer: use a Claude Desktop task, and accept that it runs when the computer is awake, not on the dot.
• It is a temporary check during work: use /loop.
• It needs your files and must run on time, every time: you have outgrown a laptop. That is what a small always-on server is for; running Hermes on a VPS is the long version.

A safe first test

Make one harmless task, such as “send me the word hello at 3pm”, in the tool you plan to use. Do not test with anything that sends, deletes or buys. Then check the result where the product keeps its history: ChatGPT’s scheduled list, or the task history in Claude, which marks skipped runs and shows why when you hover.

What working looks like. The hello arrives at the time you expect, through the channel you expect (in ChatGPT, push, email or both, set under Settings, then Notifications). If it arrives late or not at all, check the table above before you blame the prompt.