An AI agent = LLM + Tools. It doesn't just generate text. It takes action. Reads from your EHR. Queries a guidelines database. Initiates a prior auth. All from natural language prompts.

- MCP (Model Context Protocol) is what makes that possible. It's the standard that wraps APIs (and other capabilities) in a way agents can actually understand and use.

- A2A (Agent-to-Agent) is what happens next - when one agent (say, a copilot agent) needs to talk to another specialized agent (say, a clinical trial agent). A2A avoids overloading a single model and provides organizational boundaries between AI solutions.

- FHIR fills many gaps for implementing agents in healthcare. Structured data via FHIR MCP servers, patient context via SMART on FHIR, and unstructured clinical notes via Document References.

- And COIN (Conversational Interoperability)? That's the bigger vision of solving interoperability using AI standards.