The framework

The Delegation Ladder

Most advice about AI is binary — use it or don't. That framing is useless, because the same person should be on different rungs for different tasks in the same afternoon. This is the rubric the rest of the site runs on.

  1. 1

    Cede Machine

    The machine does it. You never see the work.

  2. 2

    Approve Machine-led

    The machine drafts. You review and sign.

  3. 3

    Collaborate Shared

    You and the machine work the problem together.

  4. 4

    Critique Human-led

    You do the work first, unaided. Then ask the machine to attack it.

  5. 5

    Reserve Human

    You do it alone, on purpose, knowing help exists.

Sienna and teal mark the axis: teal is machine-led, sienna is human-held. Every essay places its subject on these rungs.

How to choose a rung

Three questions, in order. They take under a minute and they resolve most cases.

  1. Can I verify the output independently and cheaply? If yes — arithmetic, a fact with a source, code with a test — rungs one and two are fine. If verifying requires the very expertise the machine replaced, you are in the verification gap and rungs one and two are not available to you, however convenient they feel.
  2. Would I have to defend this? If your name goes on it, or a person is affected by it, rung four is the floor. Form your own position, record it, then consult. The sequencing is the whole safeguard — it neutralises commission error, which is the dominant failure mode.
  3. Is this a skill I need to keep? Two tests: is it generative (does it build other capacities), and is it required to evaluate its own output? If both, budget rung-five reps on it deliberately. If neither, delegate it without guilt — most skills are in this category and pretending otherwise is just puritanism.

The inversion on rung two

Rung two looks like the safe middle and is the most dangerous place on the ladder. It is where you hold approval rights without review capacity: the machine holds the reasoning, you hold the liability. Most enterprise AI deployment lives here, and it is the crumple-zone recipe. Rung two is only honest when the three-question test passes — you have the information to disagree, the time to disagree, and the standing to disagree without penalty.

Where each topic lands

TopicRungWhy
Meaning 5 A meaning you did not construct is not yours.
Purpose 3 · 5 Choose the stake yourself; pursue it with every tool available.
Curiosity 4 Guess first, then let the machine attack the guess.
Judgment 4 · 5 Anything you would have to defend, you reach unaided first.
Responsibility 2 ⚠ Rung two is where crumple zones form: liability without capacity.
Cognitive Fitness 5 Scheduled unaided reps. Training load, not productivity.
Learning & Education 4 · 5 Attempt, then adversarial critique. Reps through the hollowed middle.
Relationships & Presence 5 Automate the gesture and you have voided the gesture.
Work & Economy 2 · 3 Where the economic value is — and where your exposure is measured.
Civilization 3 · 4 Machine-assisted analysis; human choice of what to value.

What the ladder is not

It is not a moral ranking. Rung five is not virtuous and rung one is not a failure — a great deal of what you do should be on rung one, and the productivity of rung three is real and large. The ladder exists to make the choice explicit, because the default is drift: everything slides down toward rung one, one convenient decision at a time, and nobody ever decides to let their judgment go.