Connector partially applies table creation and cannot update single-select options.
I’m using the ChatGPT Airtable connector to manage a production Airtable-backed application. The connector is useful, but table/field creation requires a lot of workaround behavior because the schema operations are extremely strict and sometimes appear to truncate or partially apply larger table creation requests.
A practical example: when creating a configuration table with multiple select fields, linked-record fields, number fields, and date/time fields, the connector created the table but only accepted part of the intended schema. I then had to add fields one-by-one, refresh schema repeatedly, handle duplicate-field collisions, and adjust linked-record value formats manually.
This is manageable for a technical user, but it makes production work slower and riskier than it needs to be. It would help a lot if the Airtable connector could provide clearer validation before applying schema changes, better partial-success reporting, more tolerant field creation helpers, and cleaner handling for linked-record fields and single-select option updates.
The underlying Airtable API may be strict, but the ChatGPT connector could smooth over more of that friction or at least return more actionable errors when a schema operation is partially applied.