The Humans Are the Bottleneck: And What We Built to Put Them Back in the Right Loop

Session Overview

Over the past two years, we've driven the cost of code generation down to nearly zero — yet our organizations have barely gotten faster. That's not a tooling problem. It's a people problem we've been refusing to face.

The real bottleneck was never typing the code. It was — and still is — the work happening around the code, and the collaboration it demands: who owns this code, who verifies it, and who's accountable when an AI agent confidently ships the wrong thing.

This talk argues that AI tools didn't create a new collaboration problem inside our organizations — they simply shone a bright light on a gap that was already there, making it impossible to keep overlooking.

What We'll Explore

• Why it's completely reasonable for non-technical teammates to distrust a flawless-looking demo
• How the unit of review is climbing — from a single line of code, to the diff, to the spec itself
• Why your test code is quietly becoming the most important spec you have in the age of AI coding

From there, we'll look at how humans and AI agents can work as one team — automating repetitive toil while deliberately pausing at the one decision that still needs a name attached to it — through a live demo of Kimchi, a service built around exactly that principle.