Great agentic AI is a data problem, we assume. Well, it's not. Not really. Data was the easy bit. The hard half is judgment: the tacit, undocumented knowing-when-to-stop your best people carry and could never put into a manual. That gap was harmless when AI just answered questions. But agents act now: book, refund, reroute, execute before anyone reviews and the same gap becomes operational risk, measured not in accuracy but blast radius. This 45 minute talk, for a mixed business and technical leadership audience, covers why the gap exists, why it matters all of a sudden, how to deliberately capture and teach judgment, and what this means for who is valuable in your organisation. The reframe: the judgment you're trying to encode is coming from your experienced people, not the role AI is replacing.