“When I became seriously ill, I moved through multiple physicians, tests, and clinical encounters. Each provider saw one part of the story. No one had a complete view of how the symptoms, history, medications, and changes in my condition connected.

This experience exposed a persistent healthcare problem. Patient data exists, but clinical teams often lack the time and tools to assemble it into a clear, longitudinal view and act before a patient deteriorates.

In this session, I will show how I used IBM Bob to turn an earlier heart failure care model into a working, governed multi-agent healthcare solution.

The original model tracked patient movement across three states: stable, deteriorated, and readmitted. IBM Bob helped transform that model into an agentic workflow with four focused agents: Patient State Monitor, Social Determinants of Health Risk Analyzer, Intervention Planner, and Care Coordinator.

The solution connects to four data sources through secure MCP servers: a calendar system for appointment and monitoring cadence, a medication tracking source for adherence and dosage changes, EHR data for clinical history and test results, and a structured spreadsheet capturing daily symptoms and medication logs kept during my own care. Each connection is scoped to the specific agent that needs it. The Patient State Monitor reads symptom and medication trends. The Care Coordinator reads calendar and EHR data to route follow-up. No agent has open access to the full data set.

The session covers the architecture, development process, human oversight, access controls, auditability, and explainability required for a healthcare agent to move beyond a demo. I will also discuss where AI should support clinical judgment and where clear boundaries must remain.

Attendees will leave with a practical pattern for designing healthcare agents that connect fragmented patient information, coordinate action across clinical workflows, and maintain the controls needed in a regulated environment.”