Skip to content

Module 01

Delegation Boundary

Sort your work by what you are willing to answer for, not by what an agent is capable of.

The question is not what an agent can do. It is what you are willing to own.

Movement IIRedesign the four decisions. Delegation, review, legibility and permission. Each is already decided by default; this is where the decision becomes deliberate and gets written down.

Capability is the wrong axis

Asked where the line is, most people answer in terms of difficulty: simple things go to the agent, hard things stay. This sorts badly. Plenty of simple work is unforgiving — a migration, a permission change, a one-line edit to billing logic — and plenty of hard work is safe to attempt badly, because the attempt is cheap and the failure is visible immediately.

The axis that actually predicts regret is not difficulty. It is what happens when the work is wrong, and how long it takes to find out.

Reversibility and detection

Two questions sort the work. If this is wrong, how hard is it to undo? And if this is wrong, how long before anyone notices?

Work that is cheap to reverse and fails loudly is nearly always safe to delegate, regardless of difficulty — the feedback loop does the supervising for you. Work that is expensive to reverse or fails silently is where your judgement is actually required, and it stays with you even when it is tedious, even when the agent would obviously succeed.

This produces a boundary you can defend to someone else, which is the test. "I don't delegate that" is a preference. "I don't delegate that because a mistake there is invisible for a quarter and expensive to unwind" is a boundary.

Method

  1. 01Take the handoffs and refusals from your baseline week.
  2. 02Score each on two axes: cost to reverse, and time to detect a fault.
  3. 03Sort into three columns — delegate, review then delegate, never.
  4. 04For every item in the never column, write the reason. If the reason is only that it feels wrong, it belongs in the middle column until you can say why.
  5. 05Look for items you refused that score as cheap and loud. Those are habits, not boundaries.

How this goes wrong

  • A boundary that is really a mood

    If the line moves with how tired you are, it is not a line. Write it down so it can be broken deliberately rather than drifted across.

  • Delegating by tool convenience

    The work an agent is best integrated with is not the work it should own. Integration is an accident of tooling; the boundary is a decision about accountability.

  • The middle column swallowing everything

    "Review then delegate" applied to all work means no boundary at all, and it will collapse under volume in module 02. Force items outward.

The exercise

  • Sort your baseline work into the three columns.
  • Write the defending reason for every item in the never column.
  • Identify two items you have been refusing out of habit, and move them.

Delegation boundary

The sections your artifact contains:

  1. Delegate — and what makes these safe
  2. Review then delegate — and what the review is checking for
  3. Never — one defended reason each
  4. Moved this week, and why

Done means. Every item in the never column has a reason that would survive a colleague asking why. Preferences are allowed, but they are labelled as preferences.

All modules

  1. 00Baseline
  2. 01Delegation BoundaryYou are here
  3. 02Review Discipline
  4. 03Estate Legibility
  5. 04Blast Radius
  6. 05Measurement
  7. 06Re-measurement