Natural-language analytics fails when the model receives table names without the business definitions that make those tables meaningful. Amazon Quick’s Agentic Catalog Experience treats that as a semantic last-mile problem.
Enterprises already maintain descriptions, keys, relationships, glossary terms, quality signals, and metric definitions in systems such as AWS Glue Data Catalog and Databricks Unity Catalog. The Quick Agent searches that catalog using the curator’s use case, surfaces relevant tables and relationships, assesses metadata readiness, and proposes a bounded set of datasets and topics.
The upstream catalog remains authoritative. Inherited table and column definitions are read-only in Quick, while primary and foreign key relationships can become preconfigured joins inside a Topic. The initial scope is deliberately narrow: richer metadata helps discovery, but only selected definitions and relationships are inherited into generated assets.
AWS’s walkthrough connects Glue for metadata and Athena for the query path. The agent explores a finance catalog, detects a star schema, then creates seven datasets and six joins in a Financial Analytics Topic. The result is immediately available for grounded questions and deterministic dashboards. Metadata refresh is manual today; scheduled synchronization is still on the roadmap.
This complements the governed context approach in WrenAI. The agent does not become the source of truth. It helps a curator turn an existing governed catalog into a smaller operational context boundary, with lineage back to the definitions and relationships that humans already maintain.