Phase Two

Phase One decided that your organization would use AI, and under what policy. Then a specific tool arrived, and the questions got harder and much more concrete.

Phase Two: A Specific Use is the implementation half of the field guide, and it is designed to be run once per application - duplicate the pages for each use, until every one of them has been named, tested, decided and recorded the same way.

Where Phase One asks whether to move, Phase Two asks what happens when you do: who is on the other end of this particular interaction and how vulnerable are they, what the tool is allowed to touch, what the contract says about your data and your exit, where to start so a failure stays bounded, and the specific observable conditions under which you stop - written down before launch, so nobody has to win an argument to invoke them.

It carries the instruments that make an implementation reviewable rather than hopeful: the use hypothesis, the stakeholder and relational-risk map, the green/yellow/red classifier applied to this use and not to "AI," the disclosure conversation, data governance and consent, vendor due diligence, pilot selection, the reversible pilot plan, the adaptive impact and risk model, a check on whether the tool is quietly hollowing out human judgment, and the implementation stop criteria.

Written for the people who will actually run the thing, and for the consultants advising them. Use it with the book, not instead of it: each instrument names its source chapter, because a checklist followed without understanding why each item is on it tends to become exactly the kind of hollow compliance exercise the book warns against.