This work is based on literally a single test. Additionally the “better” code is only evaluated on speed of processing, not accounting for readability or extensibility, which are some of the bigger issues in production code.

Yes. When I am lazy and need a quick script, I just yell at the LLM and paste error messages back into the prompt. I get to be abusive and it usually only takes a few exchanges to get a working script (simple stuff like scripts to de-duplicate a pdf library or the like)