"Generic breaks against context" sums up why so many central AI teams produce good demos and little change. The tool works in the guild's environment and fails the first time it meets a real team's exceptions and habits.

The 30 day pick up audit is a practical test. If nobody is still using an artifact a month after it was shipped, it was never the right unit of work, however good it looked in the review.

Measuring adoption instead of artifacts, and licenses versus positions, points at the same thing. Value shows up when a specific workflow in a specific team changes, not when another assistant is made available to everyone.

From the outside, this is how we try to scope client work: one team, one defined problem, and a result that can be checked after launch. Forward deployed rotations do that from inside the company. To answer your question, I think the guild survives as a hub that sets shared standards and lets the embedded engineers own the outcomes.