Airtable Omni & field agents

Omni, Airtable’s chat assistant, can build a field from a plain request like “Help our team classify incoming feedback.” The person reviewing the first answer is approving more than one answer. A field agent generates a value for every record. With Run automatically on, the agent runs again whenever its input values change, and Airtable warns that fields, formulas or automations reading the output can set off many more changes in the base.

So the real question isn’t whether the first answer looks right. The question is whether the rule should keep producing values as the data underneath changes. A plain-language build request can create standing behavior.

Airtable does protect one kind of human work. In an AI long-text field set to generate automatically, Airtable never overwrites a cell a person edited, and changing the prompt asks before regenerating existing records. A teammate’s correction is more than a value waiting to be replaced.

The same column can then hold values from an older prompt, values from the current prompt, and values a person maintains. A filled cell doesn’t say which. Airtable does track cells that are outdated or have errors, and a bulk run can target just those.

The audience matters as much as the trigger. Airtable warns that an AI field’s output can show values from source fields that are hidden from interface collaborators. A prompt telling the model to leave something out isn’t the same as keeping that field out of the inputs.

For builders: the review screen for an AI field should show what the field reads, when the field reruns, what reads the field’s output and who can see the result, before the first value is written.