The matrix sorts the tasks. Nobody asks what held them together.

Week six of an ExO 3.0 pilot produces the most important artifact in the playbook: the Task Decomposition Matrix. Every role is broken into tasks, each task is scored for agent-readiness, and the high scorers go to agents. It is the right instrument. Run carelessly, it quietly does the one thing a transformation can’t afford: it pulls apart work that only had value together.

Two lenses fix that, and both belong inside the matrix.

Where does the judgment go?

AI made production nearly free, so judgment is now the scarce resource. For every task the matrix sends to an agent, ask where the judgment that task used to carry ends up. If the answer is “nowhere,” the matrix didn’t automate a task. It deleted a decision.

What holds the bundle?

Economists Luis Garicano, Jin Li and Yanhui Wu describe jobs as bundles of tasks: some separate cleanly, some lose their value when pulled apart. Take accounts payable. Invoice matching separates cleanly. The exception call — this vendor, this dispute, this relationship — doesn’t, because the person who judges the exception is the person who answers for it next quarter.

A task can score a 5 for agent-readiness and still sit inside a strong bundle. When it does, the agent prepares and the same named person holds the call. ExO already requires this: the Fiduciary Wedge chains every agent decision to an accountable human. Separation cost is how you see where that wedge has to hold.

The same lens answers the apprenticeship question

Juniors used to earn trust on the easy work the matrix now routes to agents. A junior and an agent are the same design problem: a capable, not-yet-trusted actor placed at a seam. Pair juniors with seniors on the exceptions, and the hard cases become the training ground the easy ones used to be.

None of this changes the playbook. It adds two columns to the matrix it already runs.

The fuller arguments are in two essays: Judgment Is the Scarce Resource and What Holds the Bundle.