PATTERN Cited by 1 source
Data product as agent context layer¶
Pattern¶
Rather than exposing raw tables or unstructured documents to AI agents, wrap governed data assets in a data product layer that carries the domain context, ownership, caveats, and access semantics an agent needs to produce trustworthy answers.
Structure¶
- Catalog layer (Unity Catalog or equivalent) — tables, columns, ownership, lineage, classifications, permissions
- Product layer — ownership, lifecycle state, domain assignment, linked governed assets, rich markdown context entries
- Consumption surface — MCP server (for agents) + marketplace UI (for humans), both reading from the same product layer
Why this over alternatives¶
| Alternative | Problem |
|---|---|
| Raw table access | Agent sees schema but not business meaning, caveats, or intended use |
| RAG over documents | Disconnected from live governed data; stale; no access control |
| Schema metadata only | Missing "which business questions does this support?" context |
"Documentation becomes part of the engineering workflow, then becomes available as structured product context for the next human or agent that needs to reason over the data." (Source: sources/2026-07-21-databricks-why-rd-data-belongs-in-the-lakehouse-and-why-agents-need-it)
When to apply¶
- Enterprise data platforms where governance already exists
- Domains where trust in AI answers requires traceability to source
- Multi-team environments where domain context is local knowledge
Seen in¶
- sources/2026-07-21-databricks-why-rd-data-belongs-in-the-lakehouse-and-why-agents-need-it — cellcentric Data Hub: 27 data products with context-coverage badges, exposed via MCP.
Related¶
- concepts/data-products-as-ai-context — the concept this pattern implements
- patterns/dual-interface-human-and-agent — the dual-surface pattern often paired with this
- patterns/on-behalf-of-agent-authorization — the access-control complement