Draft table and column descriptions for the data catalog
Generate descriptions for undocumented tables and columns from their schema and data, then have owners edit and approve them.
The use case
Data catalogs are only useful if tables have descriptions, and most do not. Databricks can generate AI comments for Unity Catalog tables and columns from metadata, and BigQuery's data insights can draft descriptions for Dataplex. The owner reviews each draft, corrects business meaning, and saves it. Better descriptions also improve text-to-SQL tools that rely on them. Owners should not bulk-accept drafts; a confident wrong description is worse than none because people trust it. Teams usually start with the most-queried tables, since those descriptions get read most and help the most people.
Example workflow
- List tables with missing descriptions
- Generate draft comments
- Owner edits and approves each one
- Save to the catalog
- Re-check when schemas change
Tools
- Databricks Unity Catalog
- BigQuery
- Dataplex
Risks and guardrails
- Wrong descriptions mislead people and AI tools
- Avoid sending sensitive sample values to external models
- Assign owners so drafts get reviewed
Sources
Written by Agents Lookup. Credits and official pages for the tools this use case names:
- Databricks docs: AI-generated comments: docs.databricks.com/aws/en/comments/ai-comments
- Google Cloud docs: BigQuery data insights: docs.cloud.google.com/bigquery/docs/data-insights
- BigQuery skills (official page): github.com/google/skills (checked 2026-11-19)