Explore the critical challenges of deploying AI agents in healthcare environments through the lens of a practicing physician and AI evaluation expert. Learn why healthcare presents unique obstacles for AI agents, including clinical ambiguity, workflow interruptions, human variability, and strict risk tolerance requirements that differ significantly from other industries. Discover why traditional evaluation metrics fall short in medical settings and understand alternative measurement approaches focused on harm reduction, workflow integration, and appropriate escalation protocols. Examine how to properly scope agent autonomy to align with the distinct roles and responsibilities of healthcare professionals, including nurses, physicians, and support staff. Identify specific areas where AI agents can effectively augment clinical work while recognizing scenarios where robust human oversight remains essential for patient safety and care quality.