Agentic AI in Healthcare: Current Applications, Challenges, and Future Directions

Abstract

Agentic artificial intelligence (AI), comprising autonomous, goal-oriented systems capable of reasoning, planning, and coordinating multi-step clinical workflows under clinician oversight, is emerging as a transformative paradigm in modern healthcare. This study presents a systematic review of its conceptual foundations, architectures, applications, challenges, and future research directions. To the best of our knowledge, this is the first systematic review of agentic AI in healthcare conducted in accordance with PRISMA 2020 guidelines, providing researchers, clinicians, and policymakers with a structured, reproducible evidence base for this rapidly evolving field. A systematic search of the Web of Science and Scopus databases (January 2020–December 2025) identified 3,467 records, which were narrowed through multi-stage screening to 70 eligible studies. Results show that agentic AI systems are typically operationalized through four core components, namely planning, action, reflection, and memory, and are most commonly instantiated as multi-agent or supervisor/hierarchical architectures applied across diagnosis, clinical decision support, treatment planning, drug discovery, personalized care, and patient monitoring. Application maturity is concentrated in diagnosis and monitoring, where proof-of-concept and retrospectively validated systems predominate, while robotics and system-level applications remain largely conceptual; notably, no included study reported prospective clinical trial evidence or regulatory-cleared deployment. The most persistent barrier across architectures is the lack of standardized, transparent engineering reporting, compounded by unresolved ethical and regulatory gaps, data privacy concerns, algorithmic bias, limited interpretability, hallucinated outputs, and multi-agent coordination complexity. Addressing these barriers through robust governance frameworks, bias mitigation strategies, explainable architectures, and standardized validation protocols represents the field’s most urgent research priority. Overall, agentic AI signals a shift toward adaptive, clinician-supervised, and collaborative healthcare systems capable of delivering more efficient, personalized, and resilient care, contingent on systematically resolving these technical, evidentiary, and governance challenges.