The Self-Orchestrating Emergency Department: Agentic AI for Supervised Emergency Department Operations

The emergency department (ED) functions as a complex, high-velocity clinical environment shaped by unpredictable patient arrivals, heterogeneous acuity, competing priorities, and constrained resources. Although predictive artificial intelligence (AI) tools have shown value in selected emergency care tasks, most remain passive, providing risk estimates or decision support without coordinating system-level action. This Viewpoint argues for a shift from isolated prediction toward supervised orchestration: the use of agentic AI frameworks to support real-time management of ED operations across patient flow, resource allocation, task prioritization, and clinical workflow coordination. The concept of the self-orchestrating ED is proposed as a translational framework rather than an immediately deployable autonomous system. This model integrates three enabling components: (a) digital twins to represent the evolving operational state of the ED, (b) multi-agent reinforcement learning to support the optimization of interdependent decisions across multiple actors and constraints, and (c) large language models to facilitate communication, documentation, and context-sensitive interpretation of unstructured information.. Clinical and operational domains in which such systems may offer the greatest near-term value are outlined. Governance conditions required for safe adoption are defined, including human oversight, auditability, explainability, bias surveillance, regulatory compliance, and clear boundaries on autonomous action. Rather than presenting agentic AI as a replacement for clinical judgment, it is positioned as a supervised operational intelligence layer that could strengthen emergency care delivery if developed and implemented within rigorous technical and ethical safeguards.