01

A narrow role

Define one trigger, one purpose, permitted tasks, prohibited tasks and named owners.

02

Approved knowledge

Every service, policy, boundary and escalation rule has a source and approval state.

03

QA and UAT

Critical tests, client review and corrected retests pass before wider use.

04

Monitoring and rollback

Named people review evidence, handle incidents and can return to the prior human process.

Technical evaluator notes

Evaluation should confirm architecture, data handling, knowledge ownership, permissions, human authority, QA/UAT criteria, integration boundaries, monitoring, audit evidence, release gates and rollback for the specific workflow. The guide does not claim that external systems are connected.

Find where AI could helpBack to resources