A practitioner guide to the boundary between two fundamentally different kinds of AI work: the Creative AI Domain and the Operational AI Domain. This guidebook explains why treating all AI work as one category leads to misapplied methods, wrong risk assumptions, and avoidable failure. It introduces the Halocline concept, defines both domains, presents The Halocline Test for classifying work, and explains how failure modes, human roles, and required discipline change when work crosses the boundary. It is positioned alongside Human-Assisted AI, The Confluent Method, and The Discipline of Dependable Software as part of the same broader body of work.