実験医学における自動化からエージェンティック人工知能へ：IFCC新興技術部門の意見

Agentic artificial intelligence (AI) systems are distinguished by their ability to invoke multiple tools, compose command chains, and combine chain-of-thought reasoning with deep research to execute complex tasks and take actions. This represents a major evolution beyond machine learning and large language models (LLM), toward systems capable of planning, executing, and coordinating complex workflows. In contrast to traditional LLMs, which primarily focus on generating and classifying information, agentic AI introduces elements of autonomy, reasoning, and orchestration, while digital twins extend this concept to dynamic virtual representations of patients and laboratory processes, capable of continuous learning, simulation and adaptation. This transition has profound implications for laboratory medicine, a field characterized by high data complexity, multi-omics integration, and stringent operational demands. At the same time, laboratories face growing expectations regarding efficiency, resource stewardship, and value-based healthcare delivery. This article explores both the opportunities and limitations of agentic AI in laboratory medicine, highlighting its potential to move beyond static automation toward autonomous, outcome-driven innovation. It also examines the ethical, interpretability, and governance considerations that must accompany its implementation.