Yeah, this is the usual interaction with LLMs when coding: Ask GPT/Claude to write me a simple function. It writes 100 lines of code, trying to infer the rest of your codebase. Tell it to only solve my issue and do it in 5 lines of code. Get something close to working.