AI agent in healthcare: applications, evaluations, and future directions

With the rapid advancement of large language model (LLM) technologies, AI agents have rapidly emerged in healthcare. This review traces the historical evolution and core characteristics of AI agents, and systematically examines their applications in assisted diagnosis, clinical decision support, medical report generation, patient-facing chatbots, healthcare system management, and medical education. We further analyze existing evaluation frameworks for AI agents in healthcare, focusing on key dimensions and performance metrics. Looking ahead, we propose seven critical directions for future development: integration with embodied systems, hybrid expert models, expanded evaluation paradigms, safety and controllability assurance, ethical governance and user trust, and guidance for evolving roles of healthcare staff. This review aims to offer a comprehensive perspective on the development and implementation of AI agents in healthcare, providing theoretical support for future research, practice, and governance.