There is a version of AI consulting that leaves a team more dependent than it found them. A tool arrives, a workflow is wired to it, and the moment the consultant leaves the knowledge leaves too. The team can run the thing. It cannot change it, and it cannot explain it.
The practice is built to do the opposite, and the test for that is simple: what stays when we go.
What stays
A shared frame stays. After a keynote a room can say, in its own words, when intelligence should execute, contribute, or defer. That is not a slide; it is a way of looking at the work that survives the afternoon.
A working setup stays. After a workshop a team has one configuration built with its own hands, on its own keys, for one piece of its own work. It is small and it is theirs.
A capability stays. After an embedded engagement the people who own the work also own the layer that runs it: the rules, the examples, the sources, the loops. They can extend it, they can hand it to a new colleague, and they can switch the model under it without asking anyone.
Grounded, or it does not count
None of this works on generic material. The frame is built on the industry's own cases, the setup on the team's own brief, the layer on the decisions the team actually makes. A capability that was demonstrated on someone else's problem is a demonstration, not a capability.
That is why every engagement, whatever its size, starts from real decisions and real workflows, and why the measure of it is what the team can do on the day after.
