“Agentic AI isn’t about chatbots. It’s about systems that perceive, reason, remember, and act with humans still in the loop.”
– Alex Lee, Founder, NextTrial AI, at DevFest Boston 2025.
AI in healthcare has had many buzzwords: “generative,” “ambient,” “predictive.” But Agentic AI is shaping up to be more than the next hype cycle.
It’s a shift in how intelligence itself operates: from passive assistants to goal-driven agents that think, learn, and orchestrate complex clinical and operational workflows.
At DevFest Boston 2025, Alex Lee, one of the most articulate voices on this emerging frontier, argued that Agentic AI could become the connective tissue between healthcare’s most fragmented layers: trials, workflows, discovery, and patient engagement.
Here are Alex Lee’s insights and everything you need to know about Agentic AI in healthcare.
What exactly is Agentic AI?
In simple terms, Agentic AI is AI that acts with purpose. It doesn’t just respond to prompts.
Instead of one-off predictions or text outputs, Agentic AI systems perceive multimodal signals (like vitals, imaging, or genomics), reason about implications, remember context, and act through tools such as EHRs or regulatory databases. They improve through feedback, not retraining.
In healthcare, that means a system that can:
- Observe: read patient data, imaging, vitals, and even physician notes;
- Reason: weigh risks and simulate care pathways;
- Remember: retain context across time, not just across a single interaction;
- Use tools: query databases, clinical registries, or regulatory guidance; and
- Act: generate insights, draft notes, flag safety alerts, or even propose trial designs.
As Lee explains,
“Traditional AI models are static snapshots. They don’t adapt to evolving clinical realities or regulatory changes.
Agentic AI is dynamic. It perceives, learns, and adapts in real time. It’s what happens when intelligence starts to orchestrate instead of assist.”
That makes healthcare, with its fragmented data, evolving science, and high stakes, the perfect proving ground.
Where is Agentic AI making the biggest difference?
Here are some areas Lee mentions where Agentic AI is at play.
#1 Clinical co-pilots: Safety with scalability
One of the first places Agentic AI has found success is in clinical decision support. It is helping doctors make safer, faster, and more accurate calls.
Imagine a virtual assistant sitting inside a doctor’s digital records system. It doesn’t just type notes; it actually reviews them, spots missing details, checks against medical guidelines, and suggests next steps.
Lee points to Penda Health in Kenya as one of the earliest live deployments of an Agentic AI safety co-pilot. A GPT-4–powered assistant was embedded into their medical record system.
Over 40,000 patient visits later, the results are striking:
- 16% reduction in diagnostic errors
- 13% reduction in treatment errors
- Thousands of consultations reviewed in real time
In the U.S., major hospitals like Stanford Health and UC San Diego are working with Epic and Microsoft to embed similar copilots directly into electronic health records.
Doctors can now ask, “What changed since my patient’s last visit?” and get an instant, accurate summary.
What’s crucial, Lee emphasises, is
“These systems don’t replace clinical judgment. They reinforce it. They make human decisions safer, faster, and more explainable.”
#2 Virtual doctors: Training AI to think like clinicians
Beyond copilots lies a more daring frontier: autonomous clinical reasoning.
Lee says researchers at Tsinghua University in China have developed the Agent Hospital. It is a simulated environment where 42 “AI doctors” are being trained across 21 specialties.
Each one reasons, learns from feedback, and adapts its decision-making over time (much like a human resident).
The numbers speak volumes here, too:
- 95.6% diagnostic accuracy in controlled simulations
- 77% treatment accuracy, improving with every iteration
- Real-world pilot testing planned under human supervision
This doesn’t mean machines will replace doctors. But in remote areas or data-driven research, such “virtual clinicians” could dramatically expand access to care.
#3 Workflow & operations
Behind every consultation are hours of administrative work: notes, forms, prescriptions, follow-ups. So, Agentic AI is also being used to transform hospital efficiency.
At Chi Mei Medical Center in Taiwan, Lee says AI copilots assist doctors, nurses, and pharmacists by summarizing charts, checking drug interactions, and preparing discharge reports.
Within a month of use, the hospital saw:
- Double the pharmacist productivity.
- Documentation time cut by 60%
- 36,000 AI-assisted actions recorded
Similarly, tools like Nuance DAX (used in the U.S.) now listen to doctor-patient conversations and automatically turn them into structured notes.
#4 Patient-facing agents: 24/7 support beyond the clinic
For patients, Agentic AI shows up in the form of health companions that offer guidance, emotional support, and triage help.
- Wysa, an AI mental health app recognized by the FDA, has handled over 400 million conversations, helping users manage anxiety, depression, and chronic pain.
- Ada Health and Babylon have built AI triage systems that analyze symptoms and suggest whether you should see a doctor, self-treat, or go to the ER.
- Stanford’s Almanac AI answers medical questions using verified research sources like PubMed, offering evidence-based responses instead of guesses.
#5 Discovery agents
In labs, Agentic AI is doing something revolutionary. It is acting as a research collaborator.
At Stanford University and Chan Zuckerberg Biohub, scientists created a Virtual Lab where multiple AI agents work together to design new drugs.
In one breakthrough experiment, a multi-agent system generated 92 potential nanobodies against SARS-CoV-2 variants. And two were later validated in real lab tests.
The AI handled experiment design, analysis, and optimization while human scientists verified the results. This “human + AI” loop could cut research timelines from years to months.
“That’s AI doing what used to take entire research teams months. And it’s doing it safely, reproducibly, and explainably,” Lee notes.
Beyond co-pilots: Toward orchestrators
But the real transformation, Lee said, lies not in copilots; it’s in orchestration.
At NextTrial AI, Lee and his team are building Selena, an Agentic AI system that automates one of the most complex processes in medicine: clinical trial activation.
Traditionally, launching a clinical trial can take 120 days or more. It’s bogged down by duplicative paperwork, investigator validation, and regulatory friction.
Selena compresses that into just 30 days by automating setup, matching, and compliance with humans approving every step.
“The goal isn’t to digitize the maze,” Lee says. “It’s to delete it.”
Agents that learn, not just respond
The defining feature of Agentic AI is its ability to learn across time.
Unlike static models that reset every session, these agents build memory and context, remembering clinician preferences, patient histories, or regulatory precedents.
Lee described these agents as “living systems that learn from every trial, every patient, every clinician correction, creating a feedback flywheel that compounds value over time.”
At the post-trial stage, for example, an agentic co-pilot can automatically generate FDA submission packages, payer dossiers, and real-world evidence pipelines, while learning from previous filings to make the next faster and more compliant.
In Lee’s view, “That’s what separates automation from intelligence: the ability to remember and get better. That’s institutional memory at machine scale.”
Why it matters now
Healthcare has long been the ultimate stress test for AI: complex, fragmented, and human at its core.
Agentic AI’s promise is not just smarter algorithms but adaptive systems that can think, act, and evolve responsibly.
As Lee summed up at DevFest Boston:
“Agentic AI doesn’t replace humans. It extends them, turning every clinician, researcher, and coordinator into an orchestrator of intelligence.”
The takeaway
The AI scribe boom showed us that generative AI could save doctors time.
The Agentic AI era could go further, rebuilding trust, transparency, and adaptability into healthcare itself.
And as Lee reminds us,
“In healthcare, intelligence isn’t enough. Responsibility is the real test, and that’s where Agentic AI begins.”
Footnote:
This article draws on insights from Alex Lee’s presentation on Agentic AI delivered at DevFest Boston 2025, and the accompanying presentation deck he shared publicly on LinkedIn. Additional interpretation and analysis have been added to expand the concepts for readers.
