In 2025, we were struck by how many healthcare professionals and patients declared that they were using generative AI, despite ongoing concerns about the quality and reliability of the information being provided. We also increasingly saw the term “agentic AI,” used to describe more proactive AI systems.
Agentic AI refers to AI systems built as autonomous agents designed to pursue a goal over time. They define sub-tasks, interact with data sources and tools, and adjust their actions based on feedback. This approach builds on several prior developments, including robotics and, more recently, large language models.
In February 2025, an article in the Wall Street Journal reported on how Lumeris, a U.S. value-based care organization, deployed an AI agent to manage post-discharge follow-up.
https://www.wsj.com/articles/companies-bring-ai-agents-to-healthcare-cf9f49c1
Nearly a year has passed since that article, and there is little additional publicly documented evidence of comparable healthcare deployments.
I therefore turned to PubMed to look for peer-reviewed material on operational agentic AI. Search results from PubMed suggest that a meaningful shift occurred in 2025, when “agentic AI” began to appear as a term in the titles of articles in peer-reviewed medical journals. Despite the very small number of publications, this signals that clinicians and researchers are beginning to engage more directly with the concept of agentic AI.
Here are the three PubMed references from 2025 featuring the term “agentic AI”.
- Radiology
Agentic AI in radiology: emerging potential and unresolved challenges
The first reference is a 2025 commentary published in the British Journal of Radiology. The author, Nicolas Dietrich, describes the potential of agentic AI as follows: “Rather than awaiting specific prompts, agentic systems may independently assess imaging queues, prioritize studies based on clinical urgency, suggest additional sequences or protocols, tailor decision support based on clinical context, and dynamically adapt their outputs based on a patient’s history, prior imaging, and emerging findings.” The commentary also highlights unresolved challenges around safety, accountability, and trust, underscoring that agentic AI remains an emerging concept.
PubMed link: https://pubmed.ncbi.nlm.nih.gov/40705666/
- Orthopedics
Artificial intelligence agents in orthopedics: Concepts, capabilities and the road ahead
The European Society of Sports Traumatology, Knee Surgery and Arthroscopy (ESSKA) Artificial Intelligence Working Group published a review entitled Artificial intelligence agents in orthopedics: Concepts, capabilities and the road ahead in Knee Surgery, Sports Traumatology, Arthroscopy. The authors outline potential applications of agentic AI in workflow coordination, decision support, and longitudinal task management, while emphasizing that these systems remain largely conceptual and developmental. Clinical validation, governance, and human oversight are presented as prerequisites.
PubMed link: https://pubmed.ncbi.nlm.nih.gov/41103258/
- Biomedical and Translational Research
Talk2Biomodels: AI agent-based open-source LLM initiative for kinetic biological models
In BMC Bioinformatics, Wehling and colleagues present Talk2Biomodels, an open-source, LLM-based agentic AI platform that autonomously interacts with kinetic biological models to support research and FAIR (Findability, Accessibility, Interoperability, and Reusability) data access.
PubMed link: https://pubmed.ncbi.nlm.nih.gov/41254502/
In addition, a 2026 publication in Neural Networks introduces the MUSE (Metacognition for Unknown Situations and Environments) framework. The article highlights the importance of self-assessment and self-regulation for autonomous agents operating in unfamiliar or evolving environments. The framework does not present a direct medical application.
Metacognition for Unknown Situations and Environments (MUSE)
PubMed link: https://pubmed.ncbi.nlm.nih.gov/41046617/
Given the scarcity of publications, I also consulted ClinicalTrials.gov to identify studies in the pipeline. One trial registered in 2025 (NCT07096232) is entitled AI-Orchestrated Workflow Versus Consultant Ophthalmologist for Refractive Surgery and Keratoconus Diagnosis (AEYE Trial).
This study evaluates the performance of AEYE (Automated Evaluation for Your Eye), a multi-agent AI system designed to support ophthalmologists in diagnosing keratoconus and determining refractive surgery eligibility. AEYE simulates the clinical workflow of an anterior segment specialist by orchestrating three specialized agents: a History and Risk Agent, an Imaging Agent, and a Surgical Decision Agent.
“If successful, AEYE may offer a scalable approach to reducing diagnostic variability and improving the safety and consistency of refractive surgery screening.”
As of early 2026, Lumeris remains one of the few named, independently reported, real-world examples of agentic AI deployed in healthcare delivery. To our knowledge, no equivalent has yet been documented in peer-reviewed literature or widely reported mainstream or professional press. While agentic AI is often presented as the next step beyond generative models, documented healthcare use remains sparse: a small number of publications in 2025 mark the emergence of the term in medical literature, but real-world deployments are still the exception.