More Agents Than Employees: How Zapier Disrupted Itself Before AI Could

AI can connect any app to any other app. So why does the best model in the world still complete real business tasks only 18% of the time? In this episode of Talking AI, Matt Paige talks with Wade Foster, co-founder and CEO of Zapier — the $5 billion automation platform built on barely a million dollars raised. Foster called a company-wide “code red” the week GPT-4 launched, and he's been rethinking how work actually gets done ever since. He breaks down what Zapier's AutomationBench reveals about the gap between flashy benchmark numbers and real-world reliability, why the smartest move is blending deterministic workflows with agents instead of turning a model loose, how to tell a workflow from an agent, and why most companies have accelerated individuals without ever accelerating the company. Plus the daily AI workflow he thinks everyone is sleeping on. Watch or listen to the full episode: https://hatchworks.com/talking-ai/ -- Key Moments: 0:00 — More AI agents than employees: the setup 0:30 — Why GPT-4 triggered a company-wide code red 6:38 — Differentiation when AI is best at what you sell 9:34 — AutomationBench: the best model scores just 18.1% 11:31 — Why the top model stalls out 15:20 — Model efficiency and the token-maxing debate 19:30 — What actually counts as an “agent” 27:36 — The daily recap everyone's sleeping on 30:44 — Floor raisers vs. ceiling raisers 34:55 — Individual AI to institutional AI 37:58 — Why the six-month roadmap is dead -- Key Links: Zapier: https://zapier.com Connect with Wade on LinkedIn: https://www.linkedin.com/in/wadefoster/