Three hundred rows in, and no duplicates to clean up after

How do you get rows in without creating duplicates?

You name the fields that decide identity.

The upsert tool takes a merge list, company domain or an external id, and Airtable matches on those rather than on whatever the importer felt like comparing. Matching rows update in place, the rest are created. Three hundred rows arrive as three hundred rows.

Before any of that it reads the base schema. Field ids survive somebody renaming a column on a Tuesday, and the schema call is how it learns that a single-select will reject the value you are about to hand it.

The columns that are empty because filling them means browsing

Headcount, pricing tier, whether they have a careers page, who signed the contract.

Blank because filling them means opening a site, reading it, and typing one fact per row. Nobody does that four thousand times.

A companion does. Visit the source, extract the field, stage the record update with a link to where the value came from, then the next row. What it cannot verify stays blank and gets flagged, which is the right behaviour for a field other people filter on. This is data extraction pointed at a table you already maintain, and unlike in Google Sheets, Airtable’s typed fields reject a bad value on the way in.

Should an agent be allowed to change the schema?

It can create tables, add fields, rename and delete them, and create whole bases in a workspace. Deleting a field in Airtable takes the data in it, so a companion reads the base schema and confirms ids first, and destructive changes wait for your approval.

“Add a Status single-select with these five options, add a Last Contacted date, and backfill both from the notes column” is one instruction, and the intermediate steps are visible to you as they happen. That is how a rough base becomes a usable one in an afternoon.

Keeping a base current without watching it

Webhooks report what changed: a companion creates one on a base, then reads the pending payloads to see which records moved. Routines handle the outbound half, the weekly pass that re-checks the twenty accounts flagged for renewal and updates only the fields that moved.

For teams running their operations out of Airtable rather than a purpose-built system, that pair is what stops the base being a snapshot of the week somebody last had time to update it.

Frequently asked questions