Scheduled Tasks in Claude Cowork

1. The idea: write once, run forever

Most interactions with an AI assistant follow a familiar loop: you open a conversation, type a request, read the answer, and close the tab. That loop works well for one-off questions, but it breaks down for work that is genuinely recurring. If you ask Claude to summarize your team’s commits every morning, you have to remember to ask every morning. That is not automation — it is delegation with extra steps.

Scheduled tasks in Claude Cowork break that loop. You write a task description once — a plain-language prompt explaining exactly what you want Claude to do — and then you choose a cadence: hourly, daily, weekly, monthly, or a custom schedule expressed in cron notation. From that point forward, Cowork stores the task, triggers it at the right moment, executes it using Claude’s full computer-use capabilities, and saves the output to wherever you specified. The prompt is not just remembered; it becomes a standing instruction that drives real work on your behalf.

The accumulation effect is one of the most powerful aspects of scheduled tasks. Over time, each run adds a new output to a growing history. A weekly competitive-intelligence digest that runs for three months gives you a twelve-entry record of how a competitor’s messaging evolved. A daily standup generator gives you an accurate log of what the team shipped each day — useful for sprint retrospectives, investor updates, or simply for jogging your own memory after a vacation. The outputs compound in a way that one-off prompts never can.

2. How scheduled tasks differ from one-off requests

When you send Claude a one-off message, the context is fresh each time: you paste in the relevant documents, name the relevant people, and describe what you want. That flexibility is useful for exploratory work, but it does not scale to daily repetition. You would have to paste the same context every single morning. Humans are not good at that — we forget steps, we get busy, we introduce inconsistency.

A scheduled task solves this by treating the prompt as a template rather than a one-time message. The prompt you write contains everything Claude needs to complete the task without further input from you: which folder to look in, which URL to visit, which format to write the output in, and where to save it. Because that context is baked into the saved task, every run is consistent. Claude does not drift toward a different interpretation of your request the way a human assistant might after a few weeks.

There is an important design consequence here: scheduled tasks work best for work that does not require interactive decisions mid-flight. If your task requires Claude to ask you a clarifying question before it can proceed, scheduling is the wrong tool — an interruption defeats the purpose of automation. The ideal scheduled task is one where all decisions are already encoded in the prompt, and Claude can execute start to finish without needing to check in. This is a useful heuristic when deciding whether to schedule something: if you can describe the task completely in writing without leaving any gaps, it is a good scheduling candidate.

3. Setting up your first scheduled task

Opening Cowork and creating a scheduled task takes less than five minutes the first time. Start by navigating to the Tasks panel in the Cowork sidebar and selecting “New Scheduled Task.” You will see a prompt editor — this is where you write the standing instruction that Claude will receive every time the task runs.

Write your prompt as if you are briefing a very capable assistant who has access to your computer and no prior memory of previous conversations. Be explicit: name the folder, the URL, the app, or the document Claude should look at; specify the exact format you want the output in; tell Claude where to save or send the result. A prompt like “Summarize the three most recent pull requests merged to the main branch and save the summary to ~/standup-notes/[today’s-date].md” is complete enough to run without intervention. A prompt like “Write me a standup” is not — it leaves too many decisions unmade.

After writing the prompt, choose a cadence. The most common choices are daily (which lets you pick a specific time) and weekly (which lets you pick a day and time). For more precise control, a cron expression like 0 8 * * 1-5 runs the task at 8 AM on weekdays. Finally, specify an output destination. Cowork can save results to a local folder, append to a Markdown file, send a message to a Slack channel via the Slack integration, or open and fill a document in a running app. Once you click Save, the task is live.

4. Best use cases for scheduled tasks

Daily standups are the most popular first use of scheduled task scheduling. A well-written standup prompt tells Claude to open your git client or check the GitHub API for commits from the previous workday, pull the last few closed tickets from your project management tool, and draft a standup message in the format your team uses. Claude’s computer-use capability means it can navigate real applications — not just read text you paste in. The result is a draft standup that takes ten seconds to review and send rather than five minutes to write from memory.

Weekly reports are a step up in complexity and value. A weekly report prompt might instruct Claude to open a CRM export, calculate week-over-week changes in pipeline value, identify deals that moved stages, and write a summary with the top three trends. Sent every Friday at 4 PM to a shared Slack channel, this single task eliminates a weekly ritual that previously required pulling data manually and writing prose under time pressure.

Recurring data pulls suit teams that need a snapshot of external data on a regular basis. Claude can navigate to a web dashboard, pull the relevant numbers, and save them to a spreadsheet or structured file. A competitor pricing monitor, a Google Trends digest for your category, or a daily record of your site’s top search queries all fit this pattern.

Content monitoring works well for anyone who tracks a specific publication, RSS feed, or website for new items. A daily prompt that visits a URL, checks whether new content has appeared since yesterday’s run, and saves a digest of new titles and summaries keeps you informed without requiring you to manually check.

5. Reviewing task history and auditing outputs

Every scheduled task maintains a run log in the Cowork Tasks panel. Each log entry shows when the task ran, how long it took, whether it succeeded or encountered an error, and a link to the output it produced. This history serves several purposes beyond simple record-keeping.

When a task fails — because a website changed its layout, a file was moved, or a network connection was unavailable — the run log shows you exactly where execution stopped and what error Claude encountered. This makes debugging straightforward: you can see whether the issue is in the prompt (Claude misunderstood a step), the environment (a dependency changed), or the infrastructure (a transient network error). For tasks that interact with external websites or APIs, it is worth building a brief error-handling instruction into your prompt: “If the page does not load within 30 seconds, save a note saying ‘Data unavailable — retry tomorrow’ and stop.” This prevents a single failed run from leaving an empty or misleading output.

🧠 Check your understanding

5 questions · your answers are saved in this browser only

1. 1. What is the primary advantage of a scheduled task over asking Claude the same question each day manually?

   Scheduled tasks use a more powerful version of Claude The prompt is saved and executes automatically on a recurring cadence without manual intervention Scheduled tasks can access the internet but manual prompts cannot Manual prompts expire after 24 hours

   The core value of scheduled tasks is that you write the prompt once and Cowork executes it automatically on your chosen schedule, eliminating the need to remember and repeat the same request manually.

2. 2. Which of the following tasks is LEAST well-suited for scheduling?

   Summarizing the week's closed support tickets every Friday Pulling daily competitor pricing from a public website Drafting a standup based on yesterday's commits A multi-step research task that requires you to choose a direction halfway through

   Scheduled tasks work best when all decisions are encoded in the prompt and Claude can execute start to finish without input. A task requiring a mid-flight decision from you breaks the automation and is better handled as a one-off conversation.

3. 3. If your machine is asleep when a scheduled task is due to run, what happens?

   The task runs on Anthropic's servers instead The task is permanently skipped and marked as failed Cowork runs the task the next time it is open and detects the missed run The task reschedules itself to run one hour later

   Scheduled tasks require Cowork to be running locally. If the machine is asleep or Cowork is closed, the task runs the next time Cowork is open and detects that a scheduled run was missed.

4. 4. Which output format option is NOT described as available for scheduled task results?

   A local folder on your machine A Slack channel message A Markdown file A public GitHub repository commit

   Cowork can save results to local folders, append to Markdown files, or send to Slack. Automatically committing to a public GitHub repository is not a described output destination for scheduled tasks.

5. 5. What does the accumulation effect of scheduled tasks enable?

   Claude becomes smarter with each run because it learns from previous outputs Outputs build into a historical record that compounds in value over time Scheduled tasks are cheaper than one-off prompts because they reuse the same tokens You can share scheduled tasks with other Cowork users automatically

   The accumulation effect means each run adds to a growing history of outputs. A digest that runs for months becomes a valuable longitudinal record — something a one-off prompt can never produce.

Reset quiz

🛠️ Build it yourself

Follow these exact steps to reproduce it yourself

• Open Cowork and navigate to the Tasks panel in the sidebar
• Click “New Scheduled Task” to open the task editor
• Write a complete prompt: specify which git client or API to check, which project management tool to pull tickets from, and the exact standup format your team uses
• Set the cadence to “Daily” and choose a time 15-20 minutes before your team’s standup meeting
• Set the output destination to a local Markdown file in a folder you check each morning, or to a Slack channel your team uses
• Click Save and let the task run on its first scheduled occurrence
• Review the first output critically: is the format right, are the right commits included, is anything missing?
• Revise the prompt based on what you observe, then let it run for a week before relying on it fully
• Check the run log in the Tasks panel once a week to catch any errors caused by tool or layout changes