Announcing the Agentic Catalog Experience in Amazon Quick

Every company that's tried to roll out AI-powered dashboards has hit the same wall eventually. The data catalog is pristine, full of column descriptions and glossary terms someone spent months curating, and none of it actually reaches the tool where a sales manager types a question. AWS is now trying to close that gap with a new feature in Amazon Quick called the Agentic Catalog Experience, and the pitch is refreshingly narrow: stop making people recreate metadata that already exists somewhere else.

The mechanism is a chat-based agent scoped specifically to catalog work. A curator describes their job in plain language — something like needing tables for quarterly revenue and cost analysis — and the agent searches across the connected catalog using descriptions, tags, quality scores and Gold/Silver/Bronze labels to surface the handful of relevant tables out of what could be thousands. Once selected, the agent builds what AWS calls Catalog-Generated Datasets in bulk, using DirectQuery so nothing gets copied, and stamps them with a read-only 'Semantics Inherited' badge. Definitions for things like 'revenue' or 'active customer' come straight from AWS Glue Data Catalog or Databricks Unity Catalog instead of getting reinvented by whoever happens to be building the dashboard that week.

The more interesting piece is relationship inheritance. The agent detects primary and foreign key links between tables and automatically assembles them into Topics with star or snowflake schema joins already configured. In AWS's own walkthrough, connecting to a seven-table finance schema in Glue produced a fully joined 'Financial Analytics' Topic — two fact tables, five dimensions, six validated joins — in a single conversational pass. That's the kind of schema mapping that normally eats a data engineer's afternoon.

AWS is careful to frame Quick as a consumer of catalog metadata, not a rival catalog. Semantics stay read-only unless someone explicitly edits a dataset, at which point Quick warns that syncing back to the source catalog stops. Refreshing inherited metadata right now is a manual button push rather than a schedule, with automatic sync listed as a roadmap item rather than a shipped feature. Support for now covers Glue and Unity Catalog only, with Snowflake Horizon, Collibra and dbt — all catalogs AWS explicitly says customers have invested in — noticeably absent from the launch list.

What AWS is really selling here is trust decay prevention. Manually recreated semantics go stale the moment someone changes a definition upstream, and that drift is what erodes confidence in AI-generated answers over time. Tying Quick's datasets directly to the catalog of record, even with a manual sync button for now, at least gives teams a single place to fix a definition rather than chasing it across five dashboards.