AI often exposes ambiguity that already existed in governance, ownership, policy, and institutional memory.
A disappointing AI answer usually triggers the same questions: was the prompt wrong? Was retrieval incomplete? Was the model good enough?
Sometimes the system is working exactly as intended — simply reflecting an enterprise with conflicting policies, unclear ownership, drifting standards, and missing decision history.
AI doesn’t create that ambiguity. It makes it visible.
The optimization target eventually has to move from the prompt to the organization.
https://primamachina.substack.com/p/stop-optimizing-prompts-start-optimizing