Scheduled Tasks — Your 24/7 Workforce

The Difference That Changes Everything

In Module 5, you created your first scheduled task. It was a quick win — you picked a schedule from a dropdown, typed a prompt, and confirmed it would run automatically. You experienced the emotional payoff of watching the task history show a completed run you didn't manually trigger.

This module goes deeper. Because there's a meaningful difference between having scheduled tasks and designing scheduled tasks. A task that runs reliably at 6 AM every weekday and delivers consistent, high-quality output is not an accident — it's the product of deliberate architecture choices made before you click save.

Scheduled tasks are not just tasks that run later. They are autonomous work executed without a human present. That distinction changes how you think about every configuration decision.

The Five Schedule Options

Co-Work's scheduling interface offers five built-in frequencies:

• Hourly — for monitoring tasks, polling workflows, and high-frequency data collection
• Daily — for morning briefs, daily reports, overnight processing
• Weekly — for weekly summaries, content calendars, regular reviews
• Weekdays — for business workflows that should not run on weekends
• Manual — for on-demand tasks you want pre-configured but not automatically triggered

Paul (Advanced Use Cases) references more granular scheduling such as running a task at specific times or multiple times per day. This behavior may vary by app version — verify in your current installation whether multi-time scheduling is available before building workflows that depend on it.

The Hardware Dependency

This is the constraint that determines everything about how reliable your scheduled tasks are. Two requirements must be met for a scheduled task to execute:

1. Your computer must be awake (not sleeping, not shut down)
2. The Co-Work app must be open

If your computer sleeps during a scheduled task time, the task is skipped. The task does not queue and run when the computer wakes up — it simply doesn't run that scheduled cycle. On some configurations, a skipped task will auto-rerun when the computer wakes; verify this behavior in your own environment.

What This Means for Laptop Users

If your only machine is a laptop that you close and carry around, you will miss scheduled tasks regularly. The options:

• Sleep prevention: Configure your OS to prevent sleep while plugged in. This works for desktop-like setups but degrades battery on portable use.
• Accept the gaps: For non-critical tasks (weekly summaries, content drafts), missing an occasional run is acceptable.
• Dedicated machine: The most reliable solution, covered below.

The Dedicated Machine Strategy

The setup Paul and Jack Roberts describe:

• A Mac Mini M2 (~$600 at the time of writing) running continuously
• iCloud Drive set as the Co-Work parent folder, so outputs are accessible from all devices
• The Mac Mini runs Co-Work and executes all scheduled tasks
• Your laptop or other devices read the outputs via iCloud sync

The economics make sense if your scheduled tasks are delivering consistent business value. Think of it as the hardware cost of hiring a 24/7 employee — which brings us to how Jack Roberts frames the broader token spend question.

When you're deciding whether to run a complex analysis task on Opus every day, the question isn't "is this expensive?" — it's "what would this cost if a human did it?" A daily strategic summary that would cost a consultant $500/hour takes a few dollars in tokens. The ROI math is rarely close.

The Five Configuration Fields

Every scheduled task has five fields that determine whether it succeeds or fails. Treat each one with deliberate care.

1. Name

Use a descriptive, verb-led name that tells you exactly what the task does without opening it. "Daily Email Brief" is better than "Email Task". You'll have multiple scheduled tasks eventually — make each one identifiable at a glance in the task list.

2. Prompt (or Skill Reference)

This is the most important decision: reference a skill rather than writing a raw prompt. A skill-based scheduled task gets the benefit of all your skill refinements automatically. A raw prompt doesn't improve — you have to edit the scheduled task directly to change its behavior, which is error-prone. The pattern: create the skill first, then reference it in the task prompt using its trigger name or /skill-name.

3. Schedule

Match the schedule to the actual cadence of the work. If you're consuming a daily briefing every weekday morning, set Weekdays at 6:00 AM. If you check it occasionally, set Weekly or Manual. Over-scheduling wastes tokens and creates a backlog of outputs you never read.

4. Model

Community-tested guidance from Paul and Jack Roberts (not in official scheduling docs — test for your specific task):

• Sonnet: Routine tasks — email triage, file organization, daily briefings, regular summaries. Faster, lower cost.
• Opus: Complex analysis — UXR synthesis, financial analysis, strategic summaries that require reasoning. Slower, higher quality, higher cost.

5. Output Folder

Always specify an output folder. A scheduled task that runs but saves output to an unspecified location creates an archaeological problem — you'll spend time hunting for the output. Set a folder like ~/CoWork/Operations/Scheduled-Output/ and confirm it exists before saving the task.

Task History and Troubleshooting

Every scheduled task maintains a run history accessible at Settings → Scheduled Tasks → [task name] → History. Review this regularly — it's the only way to know if your tasks are actually running.

Common issues and their causes:

• Task skipped: Computer was asleep during the scheduled time. Check sleep settings.
• Task ran but output is wrong: The skill needs refinement. Update the skill, not the scheduled task prompt.
• Task ran but connector failed: Re-authenticate the connector in Settings → Connectors.
• Task never ran: Verify Co-Work app was open at the scheduled time. Check that the task is enabled (not paused).

The Complete Automation Pipeline

At this point, the full pipeline is assembled. You learned each stage across Modules 5 and 6:

1. Do it manually — run the task in chat, iterate until the output is right
2. Refine — run it three or more times, improving the prompt each time
3. Create a skill — codify the quality-controlled workflow as a named, reusable skill
4. Schedule the skill — reference the skill in a scheduled task with the right model and output folder
5. Monitor — check task history, refine the underlying skill when output quality drifts

The reason skills come before scheduling is that a skill can be improved in one place and the improvement propagates to every scheduled task that references it. Update the skill → every schedule that calls it gets the improvement automatically. This is the architecture that makes your automation library maintainable over time.

A raw prompt in a scheduled task is a dead end. A skill referenced in a scheduled task is a living document. The difference compounds over months.