The idea in one sentence
The agency line is where a person decides what AI does and what stays theirs.
The delegation spectrum
| Level | AI’s role | Who decides |
|---|---|---|
| 0. Human only | None | Me |
| 1. Tool | Does a bounded mechanical task | Me |
| 2. Collaborator | Drafts, critiques, summarizes | Me, on reviewing each output |
| 3. Delegate | Carries out a task; I review the result | Me, before it counts |
| 4. Autonomous agent | Acts; I audit afterwards | Me, retrospectively |
A higher level is not progress in itself. The question is always which level fits this task?
What stays mine
Purpose. Judgment and decisions. Values and ethics. Interpretation. Voice. Relationships. Accountability: I sign for the result, whoever or whatever produced it.
Four questions before delegating
- Is this the thinking I’m trusted for? If yes, AI can assist but must not replace it.
- Could I tell if the output were wrong? If not, I’m not delegating. I’m only hoping.
- Who is accountable if it’s wrong? The answer is never “the AI”.
- Does doing it myself build a skill I need to keep? If yes, keep enough practice that the skill doesn’t erode.
The understanding test
Could I explain and defend this without the AI in the room?
If not, I delegated the thinking, not just the work. That is the line.
Offloading versus surrender
Offloading hands over effort while keeping judgment, the way a calculator or a transcript does. It extends the mind. Surrender hands over judgment itself, and it diminishes the mind.
Use it
Take one task you gave to AI this week. Which level was it? Which level should it have been? Then walk the four questions.
Where it shows up
The agency question is part of every unit of the AI Leadership Program: what did AI do here, what stayed mine, and why?