Cowork scheduled tasks: work that runs without you
Define a task once, let it run on a schedule
After this, you'll be able to define a Cowork scheduled task with clear instructions, an output format, and a delivery target, and have it run on a schedule with no manual trigger.
Before you start
Complete Claude in Chrome: the browser extension that travels with you first; this lesson builds on your tour of Cowork capabilities and moves from in-the-moment help to work that runs on a schedule without you.
The idea
A Cowork scheduled task is three decisions made once, instructions, output format, and delivery target, that then runs on a schedule with no manual trigger. Instructions are what Claude should do, output format is how the result should look, and delivery target is where it lands when done. A daily news summary, a weekly digest, an end-of-day review can all run automatically.
Because this lives in Cowork, it is paid-only, and a free plan will not show scheduled tasks at all. It is also the no-code path: no folder or code project needed, and it can later combine with Computer Use for tasks that touch actual software.
One hard constraint: a Cowork scheduled task only fires while the Desktop app is open and your computer is awake. It does not run in the cloud the way a Claude Code Routine can, so a sleeping or powered-off machine means the task simply does not run.
The mental shift is from "I do this every morning" to "this is done before I sit down."
Here is the before and after: Before, every workday starts with fifteen minutes skimming five news sites and pasting headlines into a note. With a Cowork scheduled task you define it once ("every weekday at 7am, summarize these five sources as five bullets, save to my morning-brief note") and it is waiting when you wake.
Now try it: in Cowork, set up one simple scheduled task. Name the instructions, the output format, and where it should deliver. Start with something low-stakes, set it to run once tomorrow, and check that it appears.
A scheduled task is three decisions made once, what to do, how it should look, and where it lands, and after that it runs without you.
Try it (11 min)
Watch out for
- Leaving the output format vague ('summarize the news'); a fixed format like 'five one-sentence bullets' makes the result usable every day
- Skipping a test run; schedule it once first so you can confirm it works before trusting the recurring version
- Forgetting the delivery target; a finished task that lands nowhere you check is the same as no task
- Setting a high-stakes task to run unattended before you have watched it succeed a few times
- Assuming it runs while your computer is off; a Cowork task only fires when the Desktop app is open and your machine is awake, never on a sleeping or powered-off computer
Paste this into Claude
I want to set up my first Cowork scheduled task. Help me define it cleanly using three parts: instructions, output format, and delivery target. Here is what I want: every weekday morning at 7:00 AM, I want a short summary of the day's top technology headlines so I am informed before my first meeting. Please: 1. Write the exact instructions for the task in clear language. 2. Specify an output format (for example: five bullet points, each one sentence, with the source named). 3. Tell me what delivery options I should choose from (a saved note, a notification, an email) and recommend one. 4. Confirm the schedule (weekdays at 7:00 AM) and tell me how to set it to run just once first so I can test it. Keep it simple. This is my first scheduled task.
What good looks like
- The task is defined with all three parts: clear instructions, a specific output format, and a delivery target
- The output format is concrete (a fixed number of bullets, a length, sources named), not vague
- Claude recommends running it once as a test before committing to the recurring schedule
- The schedule (weekdays at 7:00 AM) is stated back to you clearly
- Nothing in the setup requires writing code
When this breaks
- Breaks when the output format is undefined, because Claude has to guess the shape each run, so the daily result is inconsistent and you stop relying on it.
- Breaks when there is no delivery target, because the task completes into nowhere you actually look, so the automation produces output you never see.
AI can help with this
In Cowork, type: 'Help me set up a scheduled task. I want [describe what, e.g., a weekly summary of my open to-dos] delivered [where, e.g., to a saved note] every [when]. Write the instructions, propose an output format, and set it to run once as a test first.'
You can now
You can complete the lesson outcome in a real Claude chat, Project, Artifact, Connector, Desktop, or Code surface.
- You can verify that the task is defined with all three parts: clear instructions, a specific output format, and a delivery target.
- You can verify that the output format is concrete (a fixed number of bullets, a length, sources named), not vague.
- You can verify that claude recommends running it once as a test before committing to the recurring schedule.
- You can verify that the schedule (weekdays at 7:00 AM) is stated back to you clearly.
Key takeaways
A Cowork scheduled task is three decisions made once: the instructions, the output format, and the delivery target. After that it runs on your schedule with no manual trigger, turning a recurring chore into a standing arrangement.
- Define every scheduled task with three parts: instructions, output format, and delivery target.
- Make the output format concrete so the daily result is consistent and usable.
- Run a new task once as a test before trusting the recurring schedule.
- Confirm the delivery target lands somewhere you actually check, or the output goes unseen.
Go deeper