Zapier VP of Product on Orchestrating 800+ AI Agents to Manage Everything | Chris Geoghegan | E286

What It Covers

Chris Geoghegan, VP of Product at Zapier, explains how the company runs 800 internal AI agents across product, engineering, and leadership functions, and outlines how product teams should rethink workflows, enterprise AI governance, and competitive positioning as orchestration replaces traditional automation.

Key Questions Answered

• •Agent Orchestration at Scale: Zapier operates 800 internal AI agents that trigger autonomously on schedules or real-world events, complete tasks, and report back without human initiation. To build this, treat each agent like a new hire: write a clear job description as the prompt, onboard with relevant context, and assign specific tools via MCP or Zapier actions.
• •API vs. MCP for Agent Tooling: APIs require agents to write recall logic from scratch each time. MCP provides a standardized layer where tools come pre-described with defined inputs and usage instructions, reducing errors and enabling agents to select and execute the right tool reliably across complex, multi-step workflows without manual intervention each session.
• •Competitive Moat Through Use-Case Data: When LLMs release competing automation features, Zapier's defensible advantage is its dataset of real automation use cases across 3.4 million businesses. The strategic play is surfacing that data to answer the enterprise bottleneck question: what should we automate next, rather than competing on raw agent-building capability alone.
• •AI Governance Framework for Enterprise: IT and security teams evaluate AI tools on three criteria: who is sending data, what data is being sent, and where it is going. Product teams selling into enterprise should build observability dashboards and access controls that answer these three questions explicitly, framing them as AI governance rather than a compliance checkbox.
• •Adoption vs. Transformation Distinction: Adoption means using AI to execute existing processes faster. Transformation means doing something previously impossible or unfundable. When measuring enterprise AI value, track both separately. Start clients on adoption metrics like time saved per workflow, then build toward transformation metrics tied to net-new capabilities the organization could not execute before AI.