The metrics look good: Velocity is up, pull requests are merging faster, and the tests are green. So why does everyone hesitate before touching a feature that's only six months old?

AI-assisted development has created a new kind of technical debt. It rarely shows up as failing builds or missed sprint goals. Instead, it builds quietly until a team can no longer explain how the system works, let alone change it with confidence.

This talk examines three forms of that debt:

* Cognitive debt grows when developers review code faster than they can understand it.
* Verification debt appears when passing tests stand in for actual confidence in the system.
* Architectural debt accumulates when AI tools produce locally sensible code that gradually pulls the larger design apart.

Recent data points in the same direction: trust in AI tools is falling even as adoption rises, enterprise security vulnerabilities have increased sharply, and experienced developers increasingly report lower productivity. We may have made fast work faster while slowing thoughtful work. And we've been measuring only the first half.

You will leave with a practical framework for spotting these forms of debt on your team, along with engineering practices for managing them. We'll look at merge standards, review habits, and "cognitive metrics" that help teams move quickly without losing their ability to reason about the systems they own.