PRs sitting unreviewed for days. Reviewers skimming 500-line diffs. Rubber-stamp approvals on changes nobody fully understood. Most engineering teams quietly accepted that PR reviews were a known weak point as the cost of moving fast.

Then AI coding agents arrived and made it unsustainable, flooding review queues with high-volume, low-quality PRs.

The fix isn't a single solution. It's layers.

This talk introduces the Swiss cheese model as a framework for thinking about PR quality: stack enough imperfect defensive layers and you get a system more reliable than any individual component. The emerging layers (spec-driven development, multi-agent competition, automated verification, and giving agents the right runtime context from the start, etc.) each address a specific failure mode. None of them are mutually exclusive.

What you'll leave with: a concrete mental model for where your current review process is failing and what a layered verification system looks like.