Quick answer
Use the agentic catalog flow when a supported upstream catalog already owns approved descriptions and relationships and you want focused DirectQuery representations that can sync those semantics. Use manual or custom enrichment when the Dataset needs transformations or Quick-specific definitions that do not belong upstream.
The decision is about ownership and lifecycle, not whether AI or a human can type descriptions faster.
| Dimension | Agentic catalog representation | Manual or custom Dataset |
|---|---|---|
| Metadata authority | Upstream catalog | Quick asset owner |
| Creation | Agent-assisted discovery and bulk creation | Author builds and documents assets |
| Query mode | DirectQuery by default | DirectQuery or SPICE |
| Inherited semantics | Read-only with manual sync | Locally editable |
| Identity propagation | Available only while DirectQuery requirements hold | Must be separately designed; Quick RLS and CLS may apply |
| Review | Tables, descriptions, inferred relationships, Topic | Transformations, definitions, joins, refresh, permissions |
Choose the catalog flow when
- Glue or Unity Catalog is the approved metadata authority;
- table and column descriptions are curated upstream;
- primary and foreign key relationships are useful for a Topic;
- DirectQuery meets performance and governance needs;
- curators want a bounded context from a large catalog;
- manual semantic sync fits the operating process.
Follow the agentic setup guide and keep the feature’s preview status explicit.
Choose manual enrichment when
- transformations are required before analysis;
- SPICE is required and approved;
- business semantics intentionally belong only to one Quick use case;
- the supported catalogs do not contain the required metadata;
- a custom calculation, join, or field must diverge from the source representation;
- the team accepts ownership of refresh, permissions, lineage, and drift.
Manual does not mean uncontrolled. Store definitions, owners, tests, and approvals in a reviewable system, and reconcile important metrics against the source.
Use a hybrid model deliberately
Begin with Catalog-Generated Datasets for governed source representations. If one asset needs customization, treat the edit as a fork. AWS states that transformations or SPICE conversion end semantic sync and make the asset a standard Quick Dataset.
Record:
- why the fork exists;
- which upstream version it came from;
- which definitions and transformations now differ;
- who owns access controls and refresh;
- how upstream changes are reconciled;
- which tests prevent metric drift.
Read the semantic inheritance guide for the lifecycle boundary.
Compare evidence, not setup time
AWS’s launch post presents a faster guided workflow, but do not promise a fixed time saving. Measure in your environment:
- curator time to an approved Topic;
- number of assets reviewed and rejected;
- relationship and metric defects found;
- permission-test failures;
- semantic drift incidents;
- end-user Q&A and dashboard correctness;
- ongoing maintenance effort.
The winning path is the one with a clear authority model and lower total cost of correct, governed analytics.
Review remains mandatory
The Agent can search catalog metadata and infer relevance or relationships. It cannot approve a table for a regulated use case, establish business ownership, or prove a join correct. Authors must review all three before sharing.
Frequently asked questions
When should I use the Amazon Quick agentic catalog flow?
Use it when approved semantics and relationships already exist in a supported upstream catalog and curators need a focused Quick representation with DirectQuery and sync.
When is manual enrichment still appropriate?
Use manual or custom Datasets when required transformations or local semantics do not exist upstream, while assigning clear ownership for the fork and its permissions, refresh, lineage, and drift.
Does the agentic flow remove curator work?
No. It can accelerate discovery and creation, but authors must review tables, descriptions, relationships, permissions, and validated query results.
Official sources
- Amazon Quick: When to use data catalog integration
- Amazon Quick: Behavior of created datasets
- Amazon Quick: The agentic catalog experience
Source check: August 4, 2026. Verify preview behavior, sync, DirectQuery, identity, and customization controls before choosing a path.