RAG works well when the answer lives in documents, but enterprise decisions rarely do. Business meaning is spread across tables, governed metrics, semantic models, ontologies, policies, and unstructured content, and an agent that sees only retrieved text can still give a confident but business-wrong answer. This session presents an architecture for grounding AI agents in a governed enterprise context layer instead of raw data alone. Using a realistic analytics-to-action scenario, I will show how to combine structured data, trusted metrics, ontology relationships, document retrieval, identity-aware access, and tool calling so the agent can understand what a business term means, retrieve the right evidence, and act within defined boundaries. We will examine routing patterns for different data types, semantic disambiguation, policy enforcement, lineage, and evaluation of grounded answers. Attendees will leave with a reference architecture and implementation checklist for building enterprise agents that can reason across data and documents while preserving business meaning, security, and auditability.

RAG retrieves text. Enterprise agents need business meaning.

That business meaning may live in:

metrics + tables + semantic models + ontology relationships + policies + documents + lineage + permissions