I see code written by LLMs like code written by a sloppy (junior) dev.

The sloppy dev will code thing twice as fast than an experienced dev, it may even be shipped. Then things will break badly in production, and it’ll take an ungodly amount of time for people to debug and untangle that messy code. In the best cases, that code is quickly thrown away and rewritten by a less sloppy dev, and the code will start working properly. In the worse case, people would add workarounds to make things work, kind of, it’ll become worse over time, no one would ever want to touch the software again.

It’s fairly questionable whether LLM code could ever be considered ‘Clean Room’ since it’s trained on, and therefore contaminated with, absolutely everything that isn’t nailed down.