Clinical AI is moving from decision support to autonomous action, with agents able to execute tasks on behalf of clinicians. The Part 2 article examines how much autonomy is appropriate for agentic AI in healthcare, weighing benefits in speed against safety and accountability. As AI can initiate orders, modify care plans, or interface with subsystems, health systems must define clear boundaries that preserve clinician oversight and the ability to intervene. The piece builds on Part 1, which traced AI’s advance beyond notes and routine workflows.
Key considerations include governance, accountability, risk mitigation, and auditability; establishing guardrails, human-in-the-loop requirements, and standards for when autonomous actions are permissible. Challenges include patient safety, liability, and data privacy. Leaders must decide which tasks may be delegated to AI agents and which demand physician control, creating scalable policies, training, and monitoring to balance efficiency gains with clinical responsibility.
