From prediction to action: a comprehensive survey on agentic AI in healthcare, foundation, application and challenges

Purpose This survey aims to examine the paradigm shift from predictive, assistive artificial intelligence (AI) to agentic AI in healthcare. Agentic AI refers to systems capable of perceiving, reasoning, planning and acting with adaptive autonomy, enabling more proactive and intelligent decision-making within complex healthcare environments. It establishes a clear conceptual foundation, systematically maps key application domains and identifies the governance, ethical and regulatory requirements necessary for the safe, responsible and effective adoption of agentic AI in healthcare systems. Design/methodology/approach A structured review (2017–2025) was conducted across major academic databases using predefined inclusion and exclusion criteria focused on autonomy, proactivity, adaptability and explainability. Findings were organized into a taxonomy contrasting passive and agentic AI, with analysis of enabling technologies, including reinforcement learning, multi-agent systems, large language models and digital twins, alongside ethical, safety and regulatory considerations. Findings Agentic AI is evolving from intelligent assistants to autonomous collaborators in diagnostics, treatment planning, surgical robotics, remote monitoring, hospital operations, patient engagement and public health. Benefits include earlier detection, individualized therapies, reduced clinician workload and improved system efficiency. Challenges remain regarding reliability, transparency, bias, privacy, accountability and workflow integration, requiring safeguards such as explainability, human-in-the-loop oversight, auditing and post-deployment monitoring. Research limitations/implications Evidence remains concentrated in pilots and simulations, highlighting the need for large-scale validation, interoperability, adaptive autonomy, fairness-aware learning and resource-efficient deployment. Practical implications The taxonomy provides guidance to developers, academia and regulators on designing and governing agentic AI solutions aligned with clinical standards. Social implications Responsible adoption can expand healthcare access, improve outcomes and reduce inefficiencies, while poor governance risks deepening inequities and eroding trust. Originality/value This work is among the first comprehensive surveys of agentic AI in healthcare. It defines what makes AI agentic, integrates applications with enabling technologies and links them to ethical and governance frameworks, reframing healthcare AI from prediction to action.