Developers now estimate that AI generates or assists 65% of the code they commit, up from 42% a year ago. And the share expecting to run autonomous software factories is set to jump from 5% to 34% within a year. But 76% say they generate AI code faster than they can verify it, and 53% spend more time reviewing code than creating it, up from 18% before AI.

Most discussion of AI coding still centers on adoption and speed. Those gains are real, but they aren’t where teams get stuck, and the costs are growing. This talk presents the second annual State of Code survey to show where the verification and governance bottlenecks bite, and what separates the teams pulling ahead. Their edge isn’t a better model or a bigger review team. It’s how they guide agents before they write code and how they verify changes after.