Also, the assumption that you can do ___ thing (tests, some dumb agent framework, some prompting trick), and suddenly magically all of the problems with LLMs vanish, is very wrong and very common.