I've been annoyed with AI-generated code so many times I don't even bother any more. It's basically: 1. Feed an idea. 2. Be given someone's mangled github repo that's somewhat similar to your idea. 3. Recognise a deep problem with the code, assuming it'll run without corrections. 4. Realise you can either correct that code and work around its weak points, or write it yourself after being made aware of easy fuckups. 5. Notice how it took more time to do all of that than to write a design document draft. 6. "Huh, turns out I can do that by feeding ffmpeg/awk/LabView/establishedToolOfChoice output into these 10 lines of Python." Something something and an email with a list of fallacies and a quote from Yudkovsky's libertarian refutations on AInobelievers. Begone! I'm done wasting time on midwits with neologism addiction. I like how llms can usually tighten my documentation, tho.