If your AI coding assistant is great at React but keeps hallucinating on your Django ORM, or fine with Python but useless on your Rust lifetimes, or solid with webdev but writes broken Godot scripts, this post is for you. The pattern is consistent across domains: general-purpose AI coding tools have a frontier they do well inside, and everything outside that frontier gets progressively worse. This post is about why, what to do about it, and when staying with a general tool is still the right call. The symptom You ask a general AI assistant to write code for a framework or engine that is not in the top 20 most-discussed on Stack Overflow. The code looks plausible. It compiles sometimes. When it runs, it breaks in ways that make you question your sanity because the error message points somewhere unrelated to the actual bug. You paste the error back. The AI apologizes and writes a different wrong version. Forty minutes in, you give up and do it yourself.…