Healthcare organizations are rapidly adopting generative and agentic AI, but evaluating these systems requires more than measuring answer accuracy. In healthcare, value also depends on whether AI retrieves authoritative evidence, preserves provenance, supports verification, and reduces the effort required to find trustworthy information.
This session presents a practical framework for evaluating evidence-grounded agentic AI across retrieval relevance, source authority, citation fidelity, provenance completeness, evidence coverage, response traceability, and search-effort reduction. Using a working healthcare discovery prototype, we demonstrate how separating query planning, retrieval, evidence ranking, and grounded response generation makes each stage measurable and auditable. The session explores how healthcare organizations can move beyond model-centric metrics and evaluate whether AI systems create real informational, operational, and clinical-support value.