A working vocabulary for 'AI strategy' that isn't embarrassing.

Every consulting deck on AI right now is built from the same seven words, deployed in the same order. "Transformative." "Augmented." "Intelligent." "Ecosystem." "Capability." "Roadmap." "Adoption." The decks are interchangeable. The recommendations are interchangeable. The decks are also, mostly, wrong — not because the recommendations are bad but because the vocabulary cannot describe the actual problem the client has.

We've been working on a different one. Not because we like jargon less than other people, but because we needed words that survive contact with a real decision. A few that have earned their keep:

Apparatus — the surrounding system around the model. Where most of the actual work and most of the actual cost lives.

Refusal — a model declining to answer because the answer would be wrong. A product feature, not a bug.

Continuity — a model's ability to remember across sessions, decisions, and corrections. The variable that separates a tool from a teammate.

Audit trail — the record of how an answer was generated. The thing that makes a model usable in a setting where the answer matters.

Floor not ceiling — the move from describing what a model can do at its best to describing what it does on a tired Tuesday. The latter is the only number that ever matters.

None of these are new. They're the words the working teams are already using. The point is that "transformative" is doing none of the work and these are doing all of it.