Why does an agent fleet need a governance control plane?

An agent fleet turns three routine questions into hard ones:

• What did our agents actually do? Thousands of actions a day, across tools and tenants, by processes that do not write tidy memos.
• What did each action cost, and to whom does it attribute? Finance needs numbers that survive scrutiny, not sampled estimates.
• When we revoke something, is it actually gone? "We pushed the change" is not the same as "no agent still holds it."

CueCrux answers all three with the same machinery: signed receipts on an append-only spine (replayable, offline-verifiable), cost attribution from fleet to agent to verb to receipt with nothing estimated, and revocation as a signed pull with per-agent acks and an attestation when zero holders remain.

The alternative is reconstructing agent behaviour from logs after the fact. That works right up until the day it matters.