We don't expect humans to do a terrible job - we just expect them to facilitate the process.

If the LLM started sketching up screens and asked questions back about the intention of the software, then I am sure people would have a much better experience.