Things You're Over-Engineering in Your AI Agent (That the LLM Already Handles)

I opened my production repo and counted the lines I wrote to re-implement things the LLM already handles. The number hurts. This post is that autopsy — 340 lines of misplaced confidence.