Generation Got Cheap. Your Review Budget Didn't.

Teams adopting coding agents measure the wrong thing. Output goes up, so it looks like a win, and then the queue of unreviewed work grows, seniors become full-time reviewers, and quality erodes in a way that shows up two quarters later.

This is a talk about capacity planning when one half of the pipeline got ten times faster and the other half didn’t: what to measure instead of lines or PR count, where to spend engineering effort on automated verification so humans aren’t the throughput limit, and how to say no to velocity claims that don’t survive contact with the review queue.