Hi everyone,
In our last thread on AI + Airtable, a clear pattern came up: field agents are the AI feature people trust most, because the prompt stays the same on every record. One member even shared a neat trick, a “personal prompt” field per user that the agent pulls in through a lookup.
That got me curious about how others set this up as their bases grow:
- Do you write one master prompt per field agent, or let teams customize them?
- Where do you keep your prompts: in the field itself, in a separate “Prompts” table, or somewhere outside Airtable?
- How do you test a prompt change before it rewrites hundreds of records?
- Have you had an agent drift or start giving worse output over time? What caused it?
We build these setups for SMBs and still don’t think there’s one right answer, so I’d love to compare notes. Happy to share what’s worked and broken for us in the replies.
Charlie, Singular (Airtable Builder Crew for SMBs)