We built a state-of-the-art RAG system for code review. In Qodo 2.4, we took most of it out.

A year ago, the smartest thing we could do for code review was index everything. Today the smartest thing we can do is index almost nothing, and remember the right things instead. I want to walk through how that happened, because the interesting part isn’t the conclusion, but the measurements that forced it, and the .