MCP and Agentic Permissions
Airtable • AI agents • New York
Impact
A company priority and cornerstone of Airtable's agent-ready platform story for their fall launch. 5+ of our largest clients are in the permissions beta with early testing showing OAuth acceptance rates increasing by over 25 points.
Overview
How might we make Airtable the platform enterprises trust to give AI agents access, without security concerns blocking the deal?
Airtable's MCP platform had already reached 75k+ weekly active users from third-party agents like Claude and ChatGPT, but permissions were still all-or-nothing at the base level. Enterprises across media, retail, and legal flagged this as a blocker, wanting to give agents narrow, specific access (e.g. read this table but not that field, or don't show any PII), and that level of control simply didn't exist yet.
I led design end-to-end: breaking a 30-page technical spec into 4 shippable milestones, using AI research agents to understand customer and usage data for the most-pressing problems to solve, and prototyping rapidly across Figma and Claude design to test table, field, and row-level permission models. I validated the designs through internal dogfooding with account teams, then shipped the frontend changes myself in Claude Code to keep pace with engineering. This feature is now in beta with several of Airtable's largest enterprise clients.
Team & timeline