Scholarlink

<p>The transition from reactive, task-specific artificial intelligence (AI) toward persistent, goal-directed agentic systems represents one of the most consequential shifts in the healthcare sector. While conventional medical AI has demonstrated measurable value in well-defined tasks, it demonstrates a fundamental inability to reason across extended time horizons and adapt to novel situational contexts in dynamic healthcare environments. Agentic AI systems, characterized by autonomous perception-reasoning-action cycles, persistent memory architectures, and integrated tool-use capabilities, offer a qualitatively different class of computational partner for clinical workflows. This review presents a comprehensive synthesis of 153 peer-reviewed publications spanning the period from 2021 to 2026, identified through a systematic search of different sources. Bibliometric analysis revealed a compound annual growth rate of 32% in publication volume, with a dramatic acceleration to 101 publications in 2025, likely catalyzed by the emergence of advanced large language models capable of supporting agentic functionality. Single-agent and multi-agent architectures demonstrated parallel and largely equivalent development trajectories reflecting an unresolved consensus regarding optimal system design for medical applications. Applications were identified across fifteen distinct clinical domains, including clinical decision support, patient engagement, precision oncology, radiology, emergency medicine, mental health, rehabilitation, pharmacy management, and healthcare operations. Across these domains, a consistent architectural scaffold emerged comprising a large language model reasoning core, layered short-term and long-term memory systems, and integrated tool access enabling real-world clinical action. Critical barriers to deployment were identified, including the persistent risk of model hallucination in zero-tolerance medical environments, the absence of standardized agentic evaluation frameworks, regulatory ambiguity surrounding autonomous clinical decision-making, and insufficient engagement with ethical dimensions of machine autonomy in patient care. This review provides a structured foundation for researchers, clinicians, and policymakers seeking to advance the responsible development and deployment of agentic AI within healthcare systems.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

From Reactive AI to Agentic Systems: A Review of Autonomous medical AI Agents in Healthcare

  • 期刊论文

  • 发表日期：

    2026年6月15日

查看全文
  • Soud Asaad Alhazba,
  • Maryam Hesham Ali,
  • Deya’ Aldeen Alawi,
  • Mohammed Azmi Al-Betar,
  • Mohamed Abd Elaziz

摘要

The transition from reactive, task-specific artificial intelligence (AI) toward persistent, goal-directed agentic systems represents one of the most consequential shifts in the healthcare sector. While conventional medical AI has demonstrated measurable value in well-defined tasks, it demonstrates a fundamental inability to reason across extended time horizons and adapt to novel situational contexts in dynamic healthcare environments. Agentic AI systems, characterized by autonomous perception-reasoning-action cycles, persistent memory architectures, and integrated tool-use capabilities, offer a qualitatively different class of computational partner for clinical workflows. This review presents a comprehensive synthesis of 153 peer-reviewed publications spanning the period from 2021 to 2026, identified through a systematic search of different sources. Bibliometric analysis revealed a compound annual growth rate of 32% in publication volume, with a dramatic acceleration to 101 publications in 2025, likely catalyzed by the emergence of advanced large language models capable of supporting agentic functionality. Single-agent and multi-agent architectures demonstrated parallel and largely equivalent development trajectories reflecting an unresolved consensus regarding optimal system design for medical applications. Applications were identified across fifteen distinct clinical domains, including clinical decision support, patient engagement, precision oncology, radiology, emergency medicine, mental health, rehabilitation, pharmacy management, and healthcare operations. Across these domains, a consistent architectural scaffold emerged comprising a large language model reasoning core, layered short-term and long-term memory systems, and integrated tool access enabling real-world clinical action. Critical barriers to deployment were identified, including the persistent risk of model hallucination in zero-tolerance medical environments, the absence of standardized agentic evaluation frameworks, regulatory ambiguity surrounding autonomous clinical decision-making, and insufficient engagement with ethical dimensions of machine autonomy in patient care. This review provides a structured foundation for researchers, clinicians, and policymakers seeking to advance the responsible development and deployment of agentic AI within healthcare systems.