Agentic Care Programs for Orchestrating Chronic Disease Management

Outpatient chronic disease care has 2 distinct failure points: (1) Between scheduled visits, where questions and symptoms go unanswered leaving patients alone and anxious, (2) Across the sequence of planned care steps, when limited specialist capacity leaves essential work, such as health education and financial counselling, delayed or unassigned. We demonstrate care programs, a feature of a healthcare AI agent platform that makes a chronic disease pathway a first-class object where AI agents coordinate and collaborate with humans to enable longitudinal, asynchronous shared care models. A care program consists of a patient cohort, a care team, a content library, and agents including a one-shot onboarding agent that screens cohort fit, a patient-facing conversational agent that answers unplanned inbound messages and books real appointments; scheduled staff-role agents that perform planned outreach; and handover encounter agents that conduct bounded financial counselling and structured health education sessions. Instead of messaging each other, agents coordinate through a shared, staff-authored and human-editable longitudinal state. In this paper, we demonstrate the system piloted at an acute tertiary hospital’s Specialist Outpatient Clinic. Specifically, we present a diabetes specialist outpatient pathway with four synthetic patient archetypes modelled after actual pilot patients. Specifically, for each archetype, we illustrate a layered safety design: deterministic screening of urgent messages before model use (red-flag screening); escalation to the staff member covering the relevant role (role-resolved escalation); human control of routine replies through review and staff takeover (reply review and staff hold); and isolation of synthetic-patient testing from real devices (test-patient isolation). For submission under Topic 1 "Frontier Models for Health" and Submission Track "Demonstration Papers".