With LLMs this is no longer true - the thing can vibe a great deal before anyone notices that they have 100.000 lines of code doing what a focused, human reviewed and tested 10.000 lines can do. And as this goes on, it becomes increasingly more difficult for anyone to actually dig into and fix things in the 100.000 without the help of LLMs (thus adding even more slop on the pile).