PubMed abstract / permitted excerpt

STUDY OBJECTIVE: To apply Autor's labor economics task framework to classify emergency physician tasks by automation susceptibility and map current artificial intelligence (AI) capabilities to each category. METHODS: We synthesized 6 published time-motion studies, ACGME Core Entrustable Professional Activities, and the O∗NET emergency physician task inventory into a unified list of 14 task categories. Two board-certified emergency physicians independently classified each task using Autor's 4-category framework. Current AI capabilities were mapped to each task using a 3-tier schema: Replace, Augment, or No Current Application. RESULTS: Nine tasks (64.3%) were classified as nonroutine abstract, 3 (21.4%) as routine cognitive, and 2 (14.3%) as nonroutine manual. No tasks were Routine Manual. AI replacement is concentrated in routine cognitive tasks (documentation, medical records review, emergency department operations management), which consume 20% to 40% of physician shift time. Augmentation dominates in nonroutine abstract domains. Nonroutine manual tasks show minimal AI penetration. CONCLUSION: Routine cognitive tasks consume a disproportionate share of emergency physician shift time, making them the immediate target for AI-driven workflow restructuring. Beyond this, augmentation of nonroutine abstract tasks is accelerating, warranting ongoing reassessment of automation boundaries across all task categories. As AI capabilities continue to expand, structured task-level analyses of this kind will be essential for anticipating workforce needs and informing AI implementation strategy, residency training design, and physician preparation in emergency medicine.