
Generalizing Refactoring Lessons and Hotspots

Refactoring Street Fighter with AI | Adam Tornhill
Topic 3 of 6
Published Oct 2, 2026 · Clip 21:42–27:15 (5:34)
Adam shares general lessons for refactoring large code bases with AI: ensure behavior preservation with system-level test automation (nonfunctional aspects included), and don't try to refactor everything. He recommends driving refactoring through hotspots — the small parts of the code base with most change activity — since refactoring untouched code is unnecessary risk. He also notes the economics of end-to-end testing have changed: agents don't complain about maintaining tests across many files, so the value of proper tests is much higher, and fundamentals of design and architecture are being reinforced.