I’ve worked in high-stakes, specialist contexts, defining where an AI-assisted feature should support a clinical decision and where it shouldn’t, and catching problems in testing that were already costing a team real time.