Short answer: no, there's no REST API or SQL statement that bulk-accepts the AI-generated comment suggestions you see in Catalog Explorer - that Accept/checkmark flow is UI-only, there's no endpoint behind it.
The workaround people actually use is to skip the "suggest then accept" UI feature entirely and generate + apply the comments yourself with ai_query(), which is a regular SQL function so it's fully scriptable:
1. Loop over information_schema.columns for the tables/schemas you care about.
2. For each column, call ai_query() against a serving endpoint (a pay-per-token FM like databricks-meta-llama-3-3-70b-instruct works fine for this) with a prompt that includes the table name, column name, data type, and maybe a few sample values.
3. Apply the result with ALTER TABLE catalog.schema.table ALTER COLUMN col COMMENT '<generated text>' (or COMMENT ON COLUMN ... IS '...').
That whole loop can be a single notebook/Python script driven off the catalog metadata, so you comment hundreds of columns without touching the UI. A couple of things worth building in:
- Batch it per-table and commit as you go, so a bad generation on one column doesn't block the rest.
- Sanity-check the AI output before applying (strip quotes/newlines, length-cap it) since ALTER COLUMN COMMENT will happily accept garbage.
- If you want a human-in-the-loop step without the UI, write the generated comments to a staging table first and review/approve there before the ALTER pass.
This is a known gap, not something you're missing - the Catalog Explorer feature and the ai_query()-based approach are two separate things under the hood.
