LLMs doing exactly what they are told is ironic and funny in the sense that we now have come a full circle, where programmers need, yet again, to be aware of better ways of doing things, keeping up with useful libraries and really knowing their sub-domain in order to avoid common pitfalls cause ... the LLMs (can) do exactly what they are told.
There is a difference here, where we simply can ask the LLM to do better, be more efficient or find better solutions. This yet again points to the "old" ways of googling, stackoverflow and other ways one has to optimize one's code, but this time, much easier (in certain domains, and sometimes with the context of your entire project).
This whole paradigm points to the old pattern of a new tool making our life much easier, with possible long-term trade-offs we don't really care about now.