Choosing a real user problem where an AI system creates verifiable value at acceptable cost and risk.
Product and use-case judgment happens before task framing. It examines the user need, current workflow, model capability, available evidence, verification path, operating cost, failure cost, and fallback. The aim is to avoid building a technically impressive agent around a problem that users do not have, cannot trust, or would solve more simply with deterministic software.