The false assumption here is that humans will always write better code than LLMs, which is certainly not the case for all humans nor all LLMs.