Cowork Lab · Step 6

Step 6 — Automation, make it run every week on its own

Turn one-off work into scheduled runs and event triggers. Then define three governance zones for what to delegate and what not to delegate.

Copilot Cowork Automation Scheduled runs Governance Hands-on
Published 2026-09-05

▶ What you learn — Until now, tasks ran each time you asked. Here, you make them run without being asked. And you discuss why that requires care.

This is the final step. Much of what you did repeats every week: weekly status summaries, unanswered-reply tracking, priority extraction. Prompting every time is also work.

6-1. Schedule a weekly status update

Scenario 15 · Climax

📌 What is happening

Project Northstar needs a status update posted to the channel every Friday. It takes 40 minutes every week, and busy weeks skip it.

🎯 Expected output
One scheduled task on the Automations page — every Friday at 16:00
💡 Efficiency point
This is the difference between one-time delegation and ongoing delegation. Once registered, it runs even if you forget. This is exactly the zone of "bad if skipped, annoying to do every time."
⚠️ Must include
Do not omit "get my approval before posting" at the end. Without it, a channel post can go up while you are not watching.
Run every Friday at 16:00.
Collect that week's Project Northstar emails and Teams conversations,
summarize confirmed decisions / open risks / next week's risks in under 3 lines,
and post it in the Project Northstar channel. Get my approval before posting.

✅ Result check

  • Is a scheduled task registered on the Automations page?
  • Is the cadence every Friday at 16:00?
  • Is the approval requirement included in the task definition?
  • Test once with “Run now” right after registration

6-2. Event trigger

Scenario 16

📌 What is happening

It can respond not to time but to an event. Mentions in an incident channel are rare, but missing one is risky.

🎯 Expected output
One event trigger — summary email on mention
💡 Efficiency point
If a schedule is a "fixed time," a trigger is a "fixed condition." Notifications already arrive, but without context. A trigger arrives with a summary.
When I am @mentioned in the INC channel, send me a one-line summary by email.

✅ Result check — Was one condition-based trigger added on the Automations page?

6-3. Governance — how far to delegate

This is the most important conversation in the lab — So far, participants have said “it can do this too” again and again. Spend the last 10 minutes on “so how far should we delegate?” A demo without this discussion will not be adopted.

Put each task into one of three zones.

Zone Criteria Examples from this lab
🟢 Always okay Reversible, does not go outside the organization, no harm if wrong Inbox classification, archiving, briefing, document analysis, my OneDrive drafts
🟡 Ask first Goes outside the organization, uses someone else’s time, or is annoying to undo Customer email, meeting booking, channel posting, meeting decline
🔴 Never Legally or financially binding, or irreversible Final contract terms, pricing approval, HR decisions, compensation promises

What you actually observed

  • Step 1 archive ran without asking → 🟢 (reversible)
  • Step 2 and 4 email drafts were created but not sent → 🟡
  • Step 5 meeting booking required approval → 🟡
  • Step 3 contract analysis only flagged issues and did not make negotiation decisions → did not cross into 🔴

Question for participants — “Of what you saw today, how many items could your organization classify as 🟢?”

Most organizations start with only two or three. Even those save hours per week. As trust builds, some 🟡 items move into 🟢. Do not open everything at once is the standard adoption path.

Full lab recap

What you delegated today

Step Task If a person does it After delegation
1 Triage + priorities + unanswered replies 40 min 3 min
2 Briefing from multiple sources 50 min 2 min
3 RFP clause analysis + deck + contract plain language 4 hours 8 min
4 Incident cause + customer email 2 hours 5 min
5 Calendar cleanup + onboarding 90 min 6 min
6 Weekly report automation 40 min every week 0 min

The real point is not time

More important than time saved is what was not missed.

  • §9.4 unlimited damages — Page 13 of 20. Easy to miss when rushed
  • 99.9% vs 99.82% — Two documents in different folders
  • 24 hours vs 72 hours — Contradiction between our promise and subcontractor promise
  • August unfulfilled promise — Inside meeting notes no one reopens
  • Threshold 8 — One line in a 3 AM chat

These five are not guaranteed to be found just because a person spends more time. They require remembering and comparing multiple documents at once, which is structurally hard for people.

Cowork’s value is less “it does it fast” and more “it does not miss what I would miss.”

Finishing the lab

  • You ran all Steps 1–6
  • You found toxic clauses §9.4 · §12.2
  • You connected the 99.9% vs 99.82% conflict
  • You found the incident cause (threshold 8) in the Teams channel
  • You registered a scheduled task and event trigger
  • You discussed your organization's 🟢 / 🟡 / 🔴 zones

What to try next

To adapt the sample data to your scenario — The principle is simple: “each fact exists in exactly one place.” If the answer exists in multiple places, you cannot prove the agent really read multiple sources.