이미지: METAL LAB 생성
Summary
- AWS has unveiled Agentic Catalog Experience, an AI workflow for data curators, in Amazon Quick.
- Quick Agent summarizes the catalog and uses natural-language conversation to find related tables and relationships, automatically generating datasets and topics.
- The feature is currently in preview, targeting AWS Glue Data Catalog and Databricks Unity Catalog.
- 발표 주체
- AWS (AWS ML Blog, 2026-07-31 게시)
- 기능명
- Agentic Catalog Experience in Amazon Quick
- 제공 상태
- 프리뷰
- 지원 카탈로그
- AWS Glue Data Catalog, Databricks Unity Catalog
- 핵심 구성요소
- Quick Agent — 탐색·생성·상속 작업 범위로 한정
- 산출물
- Catalog-Generated Datasets 및 Topics(상위 메타데이터 상속)
AWS has unveiled Agentic Catalog Experience in its analytics service Amazon Quick. It's an AI workflow that pulls semantic information already accumulated in upstream data catalogs into Quick, letting data curators build user-facing datasets and topics without manual work. The feature is currently available in preview, targeting AWS Glue Data Catalog and Databricks Unity Catalog.
The problem it aims to solve
AWS pointed out that while enterprise data teams have already built up sophisticated table and column descriptions, primary/foreign key relationships, glossaries, and metric definitions on platforms like AWS Glue, Databricks Unity Catalog, Snowflake Horizon, Collibra, and dbt, that context wasn't carrying through to end-user AI answers and dashboards. The blog post breaks this down into three issues: a discoverability limit, where it's hard to find approved assets among thousands of tables; semantic fragmentation, where upstream descriptions and relationships don't flow downstream, forcing curators to redefine everything from scratch; and, as a result, time-to-insight stretching from hours to weeks. The post also notes that when upstream definitions change, manually created semantic definitions in Quick go stale, producing "semantic drift." AWS framed the gap essentially as "the problem isn't upstream, it's the last mile."
What changes
The core addition is Quick Agent, which narrows the work down to discovery, generation, and inheritance. Based on the semantic context obtained from the catalog connection, it summarizes the entire catalog at a glance, uses natural-language conversation to pick out tables and relationships relevant to a given use case, and assesses metadata readiness. From there, a single conversational confirmation step auto-generates Catalog-Generated Datasets and Topics that inherit metadata from the upstream catalog. AWS states that this process requires no manual configuration, no switching between screens, and no multi-week setup.
Since the feature is still in preview, the original post does not specify whether catalog support will expand or when general availability might occur.
