Most agent rollouts don't fail because the agent can't do the work. They fail because somebody senior asks "what exactly did it do last Tuesday?" and nobody can answer.

An agent that works but can't be inspected is a liability wearing a productivity costume. The first time something goes wrong — and something will — the absence of a record turns a small error into an unbounded one. You can't scope the damage. You can't prove what didn't happen. So the safe move becomes turning the whole thing off.

A real audit trail answers four questions for every action: what happened, when, why it was triggered, and whether it can be undone. Miss any one and the record isn't much use. "Updated record #4821" tells you nothing without the trigger that caused it and the state before the change.

This is also what makes agents defensible to people who weren't in the room when you bought the tool. Finance wants to trace a number back to its source. Compliance wants evidence of controls. A new team member wants to understand why a process behaves the way it does. All three questions are answered by the same log.

The teams that scale agents furthest tend to be the ones who treated logging as a first-class requirement rather than a debugging convenience. They can say yes to the next automation because they can prove the last one behaved.

Capability gets you the pilot. Auditability gets you the rollout.