How far can Omni go?
That's the question I set out to test. Why bother, when Claude working through Airtable's MCP server already does so well? It's great, and I'm a fan. But not every organization can allow a connected outside LLM for security reasons. If you work under those limits and still want an LLM helping with your design and dev work, Omni is the one you'll be using.
What I tested
One of my most requested builds is what I call a Sync Method: a script that takes data from several sources and combines it into one operating table. For each source, rules decide whether a row creates a new record or updates an existing one, and every load is logged.
I gave Omni a real change request for one of these builds. Certain records are manually flagged with a status. When a later load overwrites a key field on a flagged record, the status should reset, and the old value should be appended to a history log so repeat events build up over time.
The tricky part is that once the new value is written, the old one is gone. So the solution has to capture it before the overwrite.
How Omni handled it
- It understood the problem. It read the base, found the right tables and fields on its own, and caught that the old value has to be captured before the write.
- It gave options with a recommendation. It laid out modifying the sync script, adding an automation, or doing both. It compared what each option covers and recommended the script change as the primary fix, with an automation as a safety net for manual edits. That's the same approach I would have taken.
- It wrote working code when given the real script. Its first code sample, written without seeing my script, was a reasonable sketch but not something I could drop in. That's expected: it didn't know my code. Once I gave it the original script, it produced a working version essentially the same as the change I had already built with Claude.
Takeaways
- Omni can plan like an architect. It identified the core problem, weighed trade-offs and made the right call.
- Give it the real code, or detailed specs. If it can correctly modify a 1,500-line script, it can write one from scratch when you give it enough detail about your tables, fields and rules.
- Review it like any teammate's work. Its changes were sound. What I'd adjust were matters of preference, like how results get reported and how much commenting it added.
For teams that can't use a connected outside LLM, Omni is a real option for this kind of work.
Has anyone else pushed Omni on scripting or multi-table architecture work? I'd like to hear where it has surprised you, good or bad.