AI as a skill
AI is a skill, not a tool
Tools change fast. Judgment lasts. Here it is in four levels, each harder than the last.
The four levels
From using to redesigning
- 01
Using
You get good results when the system behaves normally.
- 02
Evaluating
You can tell whether the output is right, and how you'd know.
- 03
Directing
You split real work into steps a system can do and you can supervise.
- 04
Redesigning
You reshape the problem so the system's failures land somewhere harmless.
At work
What each level looks like
Level 1: Using
In practice: You paste a messy dataset description into a chat model and get a clean schema you'd have spent 20 minutes on.
Where it stops: It transfers badly. Change the tool or the task and most of it was the interface.
Level 2: Evaluating
In practice: An agent summarises 40 interviews. You check 6 of them first, and find it dropped every negative outlier.
Where it stops: Checking costs time. The skill is choosing what to verify, and the cheapest check that catches the error you fear.
Level 3: Directing
In practice: Not 'handle onboarding emails' but: classify intent, fetch two policy clauses, draft only when confident, otherwise escalate.
Where it stops: No framework tells you where the human checkpoint belongs. That is judgment.
Level 4: Redesigning
In practice: A support drafting product becomes a policy lookup product. Same model, far less risk, more useful.
Where it stops: It takes having been wrong before, with consequences. Experience buys it. Reading doesn't.
The short version
Where each level is learned
Levels 1 and 2 come from content.Levels 3 and 4 come from practice
Practise levels 3 and 4
Four weeks, live on Zoom every Wednesday, starting November 18, 2026.
Super Early Bird: $100 for the first 30 seats.