The Advanced Research Projects Agency for Health is finalising selection this month of the innovation teams that will build ADVOCATE, the first agentic AI system aimed at FDA authorization for direct patient care. The program — formally Agentic AI-Enabled Cardiovascular Care Transformation — targets a 24/7 patient-facing AI agent that schedules appointments, adjusts medications, supports diet and exercise, and integrates with the patient's electronic health record and wearables. ARPA-H plans a competitive down-selection roughly twelve months in, then deployment inside partner health systems, with the full program running 39 months including FDA approval.
ADVOCATE matters because of where it sits on the regulatory map. The FDA recently relaxed clinical-decision-support guidance, allowing many advisory generative-AI tools to reach clinics without formal device clearance. ADVOCATE is the opposite bet — an autonomous agent acting on a patient between visits, designed from day one to clear the higher Class II/III device bar that comes with autonomous action and physician-supplanting recommendations. The three technical areas in the program's Innovative Solutions Opening cover the patient-facing agent, a supervisory component that monitors the agent's accuracy and safety in real time, and a health-systems track that recruits hospitals to deploy and study the tools.
Cardiovascular disease is the target for the same reason it has been the target for every previous wave of digital health: it is the largest cause of death in the United States, it is unevenly distributed across geography and income, and the standard-of-care interventions (medication titration, lifestyle counseling, follow-up scheduling) are exactly the kind of bounded, protocol-driven work that an agent with a supervisor loop can plausibly own. UnitedHealth has separately projected nearly $1 billion in AI-driven savings for 2026 and HCA Healthcare $400 million, signaling that the payer-and-provider economics are already converging on autonomous workflows.
Takeaway for learners: if you are early-career and interested in AI in medicine, the regulatory framework matters more than the model. The clinicians, biostatisticians, and software engineers who can navigate FDA Software-as-a-Medical-Device, the 21 CFR Part 820 quality system, and clinical evaluation reports are the ones who will end up building the agents that actually get to touch patients. ADVOCATE is the template — start reading it before the down-selection.