From one head to a shared standard.

Ask two experienced colleagues to review the same AI-generated draft. If one approves it and the other rejects it, what standard is each applying?

Factual accuracy matters. So does whether the advice fits this client, these constraints and this decision. A dependable workflow makes those expectations visible through approved sources, acceptance criteria, examples and clear escalation rules.

We would begin with experienced practitioners and difficult cases. Record what made them pause, why they rejected a plausible answer and who should decide the exception. Then test whether another colleague can apply that judgment consistently.

Before asking how much work an agent can do, establish what good work looks like and how your team will recognise it.