Calder Public Services: The second agent existed only because the first was never measured.

The problem

A correspondence-drafting agent was being followed by a second agent built to clean up its output.

What we did

Assembly measured correction rates per task type across both agents.

The result

The second agent existed only because the first was never measured. Retiring both and re-scoping one replaced them.

We were funding a second agent to fix the output of the first one.