← All articles
EN NLFR

Scheduled tasks: let Claude work while you sleep

Set up recurring tasks that Claude executes automatically — daily reports, data monitoring, inbox processing, and more. Here's how scheduled tasks work in Cowork.

Most AI workflows are reactive: you ask something, you get a result. Scheduled tasks flip that around. You set up a task once, and Claude executes it on a recurring schedule — daily, weekly, or at whatever interval you choose.

This means Claude can check your inbox every morning, generate a daily summary of your data, monitor a website for changes, or prepare a weekly report — all before you sit down at your desk.

How scheduled tasks work

In Cowork, you can schedule tasks from the task management interface. A scheduled task consists of:

  • A prompt — the instruction Claude will execute
  • A schedule — when and how often it should run (daily, weekly, custom cron)
  • A folder context — which project folder Claude should work in (including your CLAUDE.md)

When the scheduled time arrives, Claude opens a new session, reads your project’s CLAUDE.md, executes the prompt, and saves any output to your specified folder. You don’t need to have Claude open — it runs in the background.

Setting up your first scheduled task

Step 1: Define the task

Start with something simple and repeatable. A good first task:

Read all files in data/daily/ that were modified today.
Summarise the key changes in a brief morning report.
Save it as output/daily-report-[today's date].md

Step 2: Test it manually first

Before scheduling, run the task manually in Cowork to verify it produces the output you want. Iterate on the prompt until you’re happy with the result. This is crucial — a scheduled task that produces garbage every morning is worse than no task at all.

Step 3: Schedule it

In the Cowork task interface, create a new scheduled task:

  • Paste your tested prompt
  • Set the schedule (e.g., every weekday at 7:00)
  • Select the project folder
  • Enable it

Step 4: Monitor the first few runs

Check the output after the first 2-3 runs. Scheduled tasks can behave differently from manual runs because the data or context might vary. Adjust the prompt if needed.

Practical examples

Daily data summary

Task: Read today's sales data from data/sales/
Generate a summary: total revenue, number of orders, top 3 products.
Compare with yesterday's numbers (use data from the previous day's file).
Save as output/sales-summary-[date].md
Schedule: Every weekday at 7:30

This gives you a briefing waiting in your folder every morning. No manual work, no forgetting.

Weekly report preparation

Task: Read all files in data/ that were modified this week.
Generate a draft weekly report following the template in skills/weekly-report.md
Include: key metrics, notable changes, items that need attention.
Save as output/weekly-draft-[week-number].docx
Schedule: Every Friday at 16:00

By the time you’re ready to send your Friday report, the first draft is already done. You review, add your commentary on the numbers, and send.

Website monitoring

Task: Fetch the page at [URL].
Compare it with the previously saved version in data/website-snapshots/.
If there are meaningful changes (ignore timestamp/session differences),
save a summary of what changed to output/website-changes-[date].md
Save the current version to data/website-snapshots/ for next comparison.
Schedule: Daily at 9:00

Useful for monitoring competitor pricing, job listings, regulatory updates, or any page that matters to your business.

Inbox digest

If you’ve connected your email via a connector:

Task: Check my inbox for unread messages from the last 24 hours.
Categorise them: Urgent, Needs Response, FYI, Marketing/Spam.
Create a summary with the sender, subject, and a one-line preview of each.
Save as output/inbox-digest-[date].md
Schedule: Every weekday at 7:00

A pre-sorted overview of your inbox before you even open your mail client.

Data quality check

Task: Load data/inventory.csv
Run these checks:
- Any rows with missing required fields (SKU, name, price)?
- Any prices that are 0 or negative?
- Any duplicate SKUs?
- Any items where stock < reorder_point but no reorder_flag?
Save a report of any issues found to output/data-quality-[date].md
If no issues found, save a one-line "All clear" message.
Schedule: Daily at 6:00

A daily health check on your data that catches problems before they affect operations.

Tips for reliable scheduled tasks

Be explicit about file paths

A prompt that works interactively might fail when scheduled because Claude doesn’t have the same context. Always use explicit paths:

Good: Read data/sales/2026-03-24.csv
Bad:  Read today's sales file

For dynamic dates, describe the naming convention: “Read the most recent CSV file in data/sales/ (files are named YYYY-MM-DD.csv).”

Handle missing data gracefully

Your scheduled task will eventually run on a day when data is missing — a holiday, a system outage, a delayed export. Build this into your prompt:

If no new data files exist for today, save a note saying
"No data available for [date]" and exit without generating a report.

Keep prompts focused

One task, one prompt. Don’t create a mega-prompt that checks email, generates reports, and monitors websites. If one part fails, the whole thing fails. Instead, create separate scheduled tasks for each job.

Use your CLAUDE.md

Scheduled tasks read your project’s CLAUDE.md just like interactive sessions. If your CLAUDE.md specifies formatting preferences, language settings, or output conventions, your scheduled tasks will follow them too.

Review output regularly

It’s tempting to “set and forget” scheduled tasks. Don’t. Review the output at least weekly to catch:

  • Tasks that silently fail (empty or garbled output)
  • Output quality degradation (prompts that worked last month might need updating)
  • Tasks that are no longer useful (priorities change)

Limitations

No real-time triggers. Scheduled tasks run on a schedule, not in response to events. If you need “run this when a new file appears”, you’ll need a different approach.

Resource usage. Each scheduled task opens a Cowork session. Running too many tasks frequently can eat into your usage limits. Be strategic about scheduling.

No inter-task communication. Scheduled tasks are independent. Task A can’t pass data to Task B directly. If you need a pipeline, design each task to write output files that the next task reads.

Requires Claude Max. Scheduled tasks are a Max plan feature. They’re not available on Pro.

What’s next?