Please Hold for a Licensed Human

An automated message — "Your matter has been resolved. If you believe it requires a licensed human, please remain on the line" — could serve a law firm or a clinic. This essay asks three separate questions when an AI agent acts without an intervening professional: who had authority to offer the service, who owed a duty to the person relying on it, and which insurance policy, if any, responds. It discusses Crosby's $5 million AI liability policy, the boundary between clerical playbook work and advice a client reasonably relies on, an LLM study of reasons for leaving lipid-lowering therapy unchanged, correlated errors across clinics, Wyoming's unlicensed-practice statute, and the history of state medical boards after Flexner. It argues for a written coverage analysis before giving an agent authority to act, and for a human who can inspect, sample, stop, and answer for the system.