AI Agents in Medicine: A Glimpse Into the Future

With the increasing capabilities of AI in the last couple of years, many have speculated that its contribution to the field of medicine will be revolutionary. This began as excitement over discrete, individual tools that aid clinicians in their tasks, such as a program that helps radiologists identify cancerous lung lesions or an app that flags suspicious skin moles for patients.

But while these early tools offered isolated support, a new class of AI—capable of autonomous, goal-driven behavior—is starting to emerge. The rise of AI agents unlocks a whole new realm of possibilities within the medical sphere: tools that thrive in the chaos of multidisciplinary coordination, high-uncertainty situations, and patient variability.

What is Agentic AI?

Agentic AI is a type of artificial intelligence that can work independently to achieve complex goals, without requiring constant human supervision or feedback. Unlike traditional systems that follow pre-programmed rules or respond to simple commands, agentic AI can take an abstract, often open-ended goal—like "help diagnose this patient" or "plan a treatment course"—and determine a logical sequence of steps to accomplish it.

These agents don’t just give one-off answers. Instead, they break tasks down into smaller substeps, working through each one to arrive at a final solution. For instance, to diagnose a patient, an AI agent might analyze symptoms, suggest investigations, interpret the results, and generate a cohesive answer.

Through capabilities like "text-to-action" and “multi-agent collaboration”, these agents can also interact with external systems like hospital tools or databases, and collaborate with other AI agents to complete complex real-world tasks. For instance, after reasoning through a patient diagnosis, it would be possible for an AI agent to notify the doctor of a critical condition and book the patient in for a follow-up appointment on the hospital website.

What Agents Could Be Useful in Medicine?

The possibilities of AI agents in medicine are endless, but some possible ideas include:

Care Coordination Agent

This agent coordinates and tracks individualised care plans from admission to discharge and follow-up, advocating for the patient’s unique values and preferences. It keeps patients and families informed, translating medical terminology into plain language. Long-term, it could remain with the patient throughout their healthcare journey.

Patient Monitoring Agent

This agent continuously monitors vitals, labs, and other patient data to detect early signs of deterioration. It processes large volumes of real-time data, identifies subtle patterns, and alerts teams to emerging risks before they become critical. It operates on evidence-based protocols and helps doctors by providing updates on specific test results, ensuring continuous, high-quality care.

Clinical Support Agent

The Clinical Support Agent assists doctors by automating administrative tasks, including patient progress notes and ordering investigations. It synthesizes medical literature, guidelines and patient context to offer useful, timely recommendations, thereby streamlining workflows and reducing decision fatigue.

Multidisciplinary Team Management Agent

This agent coordinates the efforts of various healthcare professionals involved in patient care. It schedules meetings, integrates input from different specialists, and ensures smooth communication across disciplines. During team discussions, it presents concise patient summaries, highlights key issues, and suggests solutions to potential conflicts. It tracks action items and ensures all team members have up-to-date information, promoting informed decision-making and improving patient outcomes.

Rapid Responder Agent

Built for emergencies, this agent processes data in real-time to support triage and decision-making. It supports emergency procedures by displaying critical care guidelines, tracking interventions, and assisting in the prioritisation of tests and consultations. It integrates verbal updates, imaging, and vital signs to assist clinicians in making fast, accurate decisions.

Procedural Planning Agent

This agent seeks to provide guidance throughout medical procedures to improve safety and efficiency. Before a procedure, it assesses patient suitability. During the procedure, it offers real-time, step-by-step intraoperative guidance and helps navigate each procedural step. Post-procedure, it monitors for complications and ensures continuous care

Quality Management Agent

Operating at an institutional level, this agent continuously audits performance, flags adverse outcomes, and identifies systemic issues. It promotes quality improvement by tracking trends and unmet needs, enabling proactive resource allocation and equity in healthcare.

Conclusion

As the nascent field of agentic AI continues to evolve, much remains unanswered about where such a powerful reasoning tool would best be implemented in medicine. Nonetheless, its sole existence raises questions about how we as a society think about healthcare delivery – both as it currently is, and as it could be.

Questions for your consideration

• The predicted usefulness of AI in medicine is founded on the notion that medicine is a verifiable problem. A verifiable problem is one where the truth or correctness of a solution can be objectively checked using available evidence, criteria, or procedures. Do you believe medicine is entirely a verifiable problem?

• What other agents do you think would be useful in medicine?

• How should we test and validate agentic AI in medicine before deploying them widely?

• Are there certain tasks in medicine that should never be handed over to AI agents, no matter how advanced they become?

• What are some challenges and drawbacks in introducing AI agents in healthcare? (stay tuned for next time!)