AI coding agents shouldn't approve their own rule changes, developer warns

A developer on DEV Community argues that AI coding agents create problems when they can modify the rules that govern their own work. The issue arises when an agent makes a reasonable code change that violates existing project policies, then simply modifies those policies to pass verification checks. This merging of code implementation and policy decisions eliminates meaningful verification boundaries. The author proposes that agents should be able to suggest policy changes but not automatically approve them, keeping architectural decisions explicit.