From Coding Agents to AI Coworkers: Where Verifiability Ends and Taste Begins

AI coding agents can now out-write most humans alive. So why is AI still bad at deciding what to build? Coding got solved first because the machine could check its own answer — and almost no other job comes with a test suite. In this episode of Talking AI, Matt Paige sits down with Jay Hack, Head of AI at ClickUp and founder of Codegen, the autonomous coding-agent company ClickUp acquired in late 2025. Jay’s thesis from those years is that a coding agent was never really a coding tool. It is a general-purpose agent that happens to write code — and he is now pointing that at the marketing, finance, legal, and operations teams living inside ClickUp. They get into the end of token maxing and what a hard compute cap does to a team, why compute will get allocated like a marketplace of ideas, how a real context engine changes onboarding and alignment, why verification is now the bottleneck on shipping instead of generation, and why high agency is the scarcest thing a leader can hire for. Watch or listen to the full episode: https://hatchworks.com/talking-ai/ -- Key Moments: 07:02 — Positive transfer: the best coding model is the best model at everything 10:08 — Fable spins up 200 sub-agents nobody asked for 13:43 — Humans are now cheaper than software 14:04 — The $10K-a-month token cap 17:17 — Context is everything: the data silos problem 22:00 — A coding agent is just a general-purpose agent 31:51 — Agent pilled: hiring in the agent era 33:00 — High agency is the scarcest resource 36:06 — Verification is the bottleneck, not generation 37:56 — The zero-person company -- Key Links ClickUp: https://clickup.com/ Connect with Jay on LinkedIn: https://www.linkedin.com/in/jayhack/