It's not the fault of the "code editor developers." The underlying technology, GenAI, is non-deterministic. When you add input that varies widely in quality (prompts, existing code) what you describe is actually quite difficult.

> And it would be so simple to implement: Just some intelligent prompting or even static and dynamic analysis to detect re-occurring patterns in code and execution paths.

So just do it, then? Or are you just as incompentent yourself?

> doing exactly the right refactorings

Are you aware of any LLM models that can do that?