Airtable

What it is

Airtable is the relational database ops teams reach for when the work is too structured for a spreadsheet and too specific to model in a CRM. It runs at 500,000+ organizations including 80% of the Fortune 100, and Bending Spoons put its annual recurring revenue at roughly $480M as of June 2026, growing over 20% year over year.

Since the June 24, 2025 relaunch as an AI-native app platform, “Airtable AI” is three separate things, and telling them apart is what decides your bill. Omni is the builder and analyst — describe an app in a prompt and it writes the tables, fields, relationships, and interfaces. Field Agents work a single record at a time: read an attached document, search the web, translate, extract structure out of a transcript, and fire on their own whenever a row is created or updated. Field AI fills individual cells. Airtable also runs an official MCP server at https://mcp.airtable.com/mcp on every plan, over OAuth or a personal access token, so Claude can query bases, create records, and build new ones.

Then there is the thing no feature page will tell you. On August 4, 2026 Bending Spoons signed a definitive agreement to buy Airtable for $1.285B enterprise value — about $2.25B equity including net cash, all cash — against the $11.7B valuation Airtable carried in 2021. Close is expected later in 2026 pending regulatory approval, and both companies operate independently until it lands. That timeline, not the roadmap, is what you write a contract around this quarter.

Why it shows up in ops stacks

• Omni writes the schema, not just the query. You get tables, relationships, and a working interface from one prompt, which collapses the two-week “model the process first” phase that stalls most internal-tool projects. It does not build automations — you still wire those by hand.
• Field Agents are per-row, and per-row is where ops work lives. n8n and Zapier give you an event-driven flow that you own and maintain. A Field Agent is a column that fills itself. For a 400-row intake table that needs the same extraction on every row, that difference is the whole build.
• AI stopped being a per-seat surcharge. Since June 24, 2025, Team, Business, and Enterprise customers are no longer billed per user for Airtable AI. Credits come bundled with the plan instead, which makes a pilot cheap to start and expensive to misjudge.
• The MCP server ships on the free plan. No enterprise SKU, no sales call, so you can test whether an agent can actually drive your base before anyone signs anything.

Pricing

Per user per month from Airtable’s pricing page, checked September 6, 2026. Billing is per editor — read-only collaborators, form submissions, and share links cost nothing.

• Free — $0. 500 AI credits per user with Editor permission or above.
• Team — $20 annual, $24 monthly. 50,000 records per base, 15,000 AI credits per billable collaborator.
• Business — $45 annual, $54 monthly. 125,000 records per base, 20,000 AI credits per paid user.
• Enterprise Scale — custom. 25,000 AI credits per paid user at list price.

Credits are the line teams misprice. Airtable’s own billing guidance puts a simple question at about 10 credits, categorizing one piece of feedback at about 1, analyzing a 10-page contract at about 200, a 50-page document at 500 to 1,500, and a 100,000-word input at 2,000 to 5,000 or more. Building with Omni burns nothing. The meter starts when Field Agents touch rows.

Do the arithmetic before you sign. A Business seat’s 20,000 included credits is roughly 100 ten-page contract reviews a month. A legal-ops team running 500 contracts through Field Agents lands near 100,000 credits and pays $200/mo in overage on top of seats — packs run $20/mo for 10,000 credits up to $800/mo for 400,000, a flat $2 per 1,000 either way. So the real Business number for an AI-heavy team of 15 is about $8,100 a year in seats plus $2,400 in credits, not $8,100.

Superagent — launched January 27, 2026 at superagent.com, Airtable’s first standalone product in 13 years — is a separate product on a separate bill. It has been reported at $20 per user monthly at entry up to $200 for power users with bundled inference credits. Airtable has not published a Superagent price sheet, so treat that band as reported rather than contracted.

Best for

RevOps, legal-ops, recruiting, and CS teams at 10-200 seats that already keep a system of record in Airtable and want AI applied to the rows in it, on Business at $45 a seat annual. The scoped case where Airtable wins outright: a structured intake table where every new row needs the same research or extraction — vendor security reviews, inbound contract triage, sourcing lists, support-ticket enrichment — and you want that work happening inside the row rather than in a flow you maintain somewhere else.

Not for teams that want AI to build their automations, because Omni does not write them. Not for orchestration across ten systems — that is n8n or Workato, and Airtable is a node inside it, not the layer above it. Not for anyone about to sign a multi-year enterprise deal in the next two quarters.

Alternatives — and when to pick them instead

• Smartsheet — one of the two share leaders in enterprise work management. Pick it when the job is PMO reporting, resource management, and portfolio rollups, and when the deciding constraint is surviving a procurement review rather than building anything new.
• monday.com — the other incumbent by share. Pick it when the buyer is a department head who wants boards, views, and dashboards on day one and has no appetite for schema design. You give up the relational modeling that makes Airtable worth the seat price.
• Notion — pick it when the deliverable is a document with a database attached rather than a database with views attached. Wrong pick for a 100,000-row operational table.
• Teable and Baserow — the fastest-growing entrants in the segment, open-source and self-hostable with an Airtable-shaped model. Pick either when the acquisition risk below is disqualifying, when data has to stay on your own infrastructure, or when per-editor seat pricing is what broke the business case. You give up Omni and Field Agents entirely, which is the trade.
• Retool — pick it when the app is the deliverable and the data already lives in Postgres. Airtable is the wrong shape when you have a real database and need a front end for it.

Watch-outs

• The acquirer has a documented repricing pattern, and your renewal is inside its window. Bending Spoons took Evernote’s Pro plan from $37 a year to $250 a year after the 2022 acquisition, and confirmed plans to cut about 75% of a workforce of 350+ at WeTransfer within roughly two months of that 2024 deal. None of that has been announced for Airtable and the transaction has not closed. Guard: do not sign a multi-year renewal before close. Negotiate a price-lock or an uplift cap that runs through your next renewal, put it in the order form rather than the MSA, and take a full base export now so the switching cost you would face is a known number rather than a discovery.
• Credits are metered per row, and the pilot is where the number hides. A Field Agent set to run on create-or-update will process backfills and bulk imports you did not think of as AI usage. A single 300-row CSV import of long documents can clear a Business seat’s monthly allowance on its own. Guard: run the pilot on one table with the trigger set to manual, read the credit consumption after 100 real rows, multiply by your actual monthly volume, and only then switch the agent to automatic. Set the budget at the number you measured, not at the included allowance.
• Record limits are per base, not per workspace, and they bite mid-migration. Team caps at 50,000 records per base and Business at 125,000. Ops teams consolidating several trackers into one base hit the ceiling after the merge, when the rollback is most expensive. Guard: count total records across every table you intend to consolidate before you start, and if the total lands within 20% of the cap, split by entity into separate bases and link them instead of migrating first and discovering the wall.
• The MCP server does less than the marketing implies. It creates records 10 per request, and it cannot deploy automations, edit Script actions inside them, add interface filters, or modify existing interfaces. Guard: confirm your agent’s exact tool path on a free-plan base before you make Airtable an MCP dependency in a stack. If the agent needs to change automations rather than rows, that work stays in the Airtable UI and no amount of MCP wiring moves it.