Airtable Omni 2.0 review: a fast builder for messy spreadsheets, with the agent controls still arriving

The verdict, by plan

Airtable's Fall Release, announced October 21, 2026, put Omni 2.0, its built-in AI assistant, on every plan. Hand it spreadsheets, PDFs or images, however messy, and Airtable says it will build the data model, import the data and set up automations and AI fields in one go. It can also set up Field Agents that summarize, classify or enrich records, write automations from a plain-language description, and write scripts.

Our verdict depends on what you're paying for and what you want the AI to do.

Free and Team: use it to build. Turning a shared spreadsheet into a proper base is exactly the job small teams put off, and an assistant that does the first draft of the structure is a real time saver. Review the structure before you invite everyone in.

Business: use it to build, and be careful about letting agents write to live data until the controls below arrive. Audit logs aren't on this plan.

Enterprise Scale: you have the most visibility today, thanks to the new agent audit logs. Wait for agent permissions before giving agents broad write access to production bases.

Building a base is a one-time risk. Letting an agent write to it every day is a standing one.

What it does well

The strongest part of the release is the boring part: moving a team off spreadsheets. Most small businesses have a few critical sheets that grew without a plan: an inventory list with merged cells, a client tracker with five tabs of half-duplicates, a vendor list in a PDF. Omni 2.0 is pitched squarely at that. A good first-pass structure, with linked tables and sensible field types, is the hardest step for non-technical teams, and it's the step that most often stalls.

Field Agents are the other useful piece. Summarizing a long request, tagging a support message by type or enriching a record with a short description are tasks where an AI is good enough and a person reviewing a sample is cheap.

The connections got broader too. Airtable's MCP, which lets outside AI tools work with your bases, now connects with Claude, ChatGPT, Gemini and Cursor, with more promised. Admins can turn it on or off for the whole organization, approve which tools people can use, and limit each tool to the bases it needs. That's the right set of switches.

And there's Canvas, a new feature in open beta that builds custom apps, dashboards or presentations on your live data from a plain-language description. It requires the organization to enable AI, AI Labs and Anthropic as an AI provider, and it isn't yet available for HIPAA-compliant organizations or those with non-US data residency. Treat it as a preview.

Where it falls short today

The gap is control over what an agent can touch. Airtable lists two important pieces as coming soon: agent permissions down to the field or record, so an agent can work through customer feedback without seeing contact details, and the rule that an agent never gets more access than the person who connected it. Until those ship, an agent's reach is roughly the reach of whatever access it was given, and on many bases that's broad.

Visibility is uneven across plans. The new audit log tells you whether a person or an agent took an action and in which base, attributes third-party tools to the integration and the person who connected it, and exports to CSV or JSON or streams to security tools through an API. In the release thread, an Airtable community manager confirmed audit logs are Enterprise Scale only. So a 40-person company on Business can let an outside AI tool write records and not have this log to see what it did.

Bring-your-own agents and keys are also coming soon, so organizations running their own autonomous agents on Airtable are waiting too. None of this is unusual for a release that ships a lot at once. It just means the build features are ahead of the safety features. We've argued that an agent audit should test behavior, not just read logs, and that advice stands whichever plan you're on.

Pricing and the alternatives

Airtable's pricing page, as of November 1, 2026, lists a Free plan, Team at $20 per user per month and Business at $45 per user per month, both billed annually, and Enterprise Scale at custom pricing. Seats are charged for users with edit permissions. Omni 2.0 itself is on all plans, but Canvas has the organization-level requirements above, and audit logs sit on Enterprise Scale.

If what you want is AI tools acting across many apps rather than a database, two alternatives came up in the same weeks. Make introduced MakeOS on October 20, an environment where agents and actions are meant to be visible, permissioned and reversible; it's in closed beta with no published pricing. Make's assistant, Maia, gets a Plan mode where you approve changes before they run, plus audit logs for changes made by agents and MCP clients. Zapier made one Zapier MCP server usable across Claude, ChatGPT, Cursor and other tools, with a managed mode where you choose which actions each client may run and lock fields; Zapier says that's included on every plan. We've covered how Zapier moved agents into Zaps before.

The difference in short: Airtable is where your data lives, Make and Zapier are how it moves. Many businesses end up with one of each.

Who should switch it on, and how

Turn it on now if you're a small team with spreadsheets you've been meaning to tidy. Let Omni build the first draft, then have the person who knows the work review every table and field before data goes live.

Turn it on in a sandbox if you're on Business and want Field Agents or MCP access. Copy a base, let the agents run for a couple of weeks on the copy, and check a sample of their changes by hand.

Wait for permissions before giving any agent write access to a base with customer contact details, payments or anything regulated.

If you'd like help moving your team off spreadsheets, or wiring Airtable to the rest of your tools with sensible controls, our business automation team does that work. Get in touch and tell us which sheet you'd rescue first.

Related work: Sutton Place.