People who use agents well also keep a few tasks analog on purpose. Not as a purity test. So they can still tell when a report is thin. If every outline, every recap, and every first draft is delegated, your taste goes quiet.
Skill atrophy is a choice you can see.
This is not the pitchfork test from Working with AI (protect work you love). This is the calibration test: keep enough of the craft that you can still catch a bad agent.
Core principles
- Keep one recurring task in each family (write, analyze, recap) that you sometimes do without AI.
- If you cannot explain why an output is wrong, you are not ready to delegate that job.
- Delegating research does not delegate the decision. You still own the send button.
- A week of “I never opened Chat” is not a moral win. A week of only Chat on every job is not fluency.
Related: Your AI Usage — shipped vs spin
Check yourself
Why keep some work analog if agents exist?
You cannot review what you can no longer do. Calibration is the point, not nostalgia.
Do this in Copilot
Pick one task from this week you will do without Copilot, on purpose. Write why: taste, speed, or keeping the skill.
Paste into Copilot Chat. You want an argument you can disagree with.
Name what you will not delegate
Here is my brief from Lesson 1: [PASTE]. Here is a second task I might keep analog: [TASK]. Argue once for delegating it and once for doing it myself. Then stop. I will choose. Do not tell me that using more AI is always better. Do not use the word “advanced.”
Did you run this in Copilot? Mark complete when you have tried it.
RecordedNext lesson: One brief, three routes →
Navigate: press j for next lesson,k for previous.