Imagine an app on your phone that doesn’t just track your steps, but actually adjusts your heart medication dosage after analyzing your latest vitals—without a human doctor logging in. That’s the reality the Trump administration’s ARPA-H agency is betting on, and they want it to be FDA-approved in just over three years.
The News
On January 21, 2026, the Advanced Research Projects Agency for Health (ARPA-H) announced the “ADVOCATE” program (Agentic AI-Enabled Cardiovascular Care Transformation). The initiative aims to fund and develop “agentic” AI—artificial intelligence that can take independent action—specifically for managing cardiovascular disease.
Unlike today’s “predictive” tools that merely flag risks for humans to review, these AI agents are intended to autonomously schedule appointments, support diet changes, and even adjust medication dosages. The program outlines a blistering 39-month timeline to go from concept to FDA authorization, creating a regulatory fast lane for generative AI in high-risk clinical settings.
Crucially, the program demands a dual-agent architecture. It calls for a “clinical agent” that interacts directly with patients to manage their care, paired with a separate “supervisory agent” that monitors the clinical agent’s decisions in real-time. This “AI watching AI” structure is ARPA-H’s answer to the safety concerns surrounding generative models. Haider Warraich, M.D., the program manager, was explicit about the goal: “The FDA cannot authorize an AI doctor, but they can authorize a technology that improves outcomes for patients with heart failure.”
Why It Matters
This marks a pivotal shift from “AI as tool” to “AI as provider.” Until now, FDA approvals have largely been stuck in the “predictive” era—algorithms that act as second opinions for radiologists or risk scorers for clinicians. ADVOCATE is pushing for the first FDA-authorized “agentic” AI that directly interacts with patients 24/7.
The stakes are incredibly high. Heart disease kills 200,000 people annually in the U.S., often because simple interventions—like medication adjustments—happen too slowly in our overburdened system. A patient might wait months to see a cardiologist just to have their diuretic dose increased. By automating this “blocking and tackling” of routine care, the government hopes to democratize access to top-tier cardiology. It promises a future where a rural patient has continuous access to specialist-level decision-making without leaving their home.
Furthermore, this program is designed to break the regulatory deadlock. By actively funding the development of a device that sits clearly within FDA authority, ARPA-H is forcing the agency to define the rules of the road for generative AI before the market is flooded with unregulated tools operating in the “grey zone.”
The Skeptic’s View
Handing over the prescription pad to an algorithm terrifies many in medicine. John Whyte, CEO of the American Medical Association, has cautioned that mistakes in diagnosis or dosing can be life-threatening and that “clear physician oversight” is non-negotiable.
The technical hurdle is also massive. Generative AI is notorious for “hallucinations”—confidently stating false information. While the “supervisory agent” concept is theoretically sound, it is unproven at this scale. Who watches the supervisor? If the clinical agent recommends a dangerous dosage and the supervisor fails to catch it, the consequences are immediate and physical.
There is also a concern about transparency. Warraich noted that while the supervisory agent might be open-source to encourage broad safety standards, the clinical agent—the core product—will likely be proprietary. This could lead to a “black box” scenario where the logic driving life-or-death decisions is shielded from public scrutiny behind corporate trade secrets.
Looking Ahead
The clock is already ticking. ARPA-H plans to select the winning research teams by June 2026. The program will follow a rigorous “down select” process, eliminating underperforming teams after the first year.
Watch closely to see who throws their hat in the ring. Will we see partnerships between established electronic health record vendors and agile AI startups? And most importantly, will the FDA be able to evolve its review processes fast enough to meet the 39-month deadline? This isn’t just about heart disease; it’s a dry run for the entire future of automated medicine.