I believe your question answers itself, if you could do this then people would[2]. The next question might be 'What prevents the current LLM based coding tools from being able to do this?"

Now there has always been software that people have used that they "could" write for themselves but they didn't because they didn't want to maintain it. Seems like things that are like that would be good candidates for this right?

But this question shares at least a spiritual geneology with the response to some new startup code that does some thing, and the comment here is "Gee, I can't believe that got this funded, I could write that in a weekend."

The problem is, there are many things that are easy to describe but difficult to pin down, and reality is complicated. So many times I've gone into a project thinking it would be "easy" only to start uncovering layer after layer that are much more involved than I expected.

So no, you cannot '1-shot prompt' pretty much any significant software, and that goes doubly true if it affects your life in any potentially devastating way.

One of the things I'm constantly amused by[1] is the Solitaire program on iOS that is clearly being vibe coded these days because it comes up with daily challenges like 'win a hand in -2739551 moves'. Like no human would ever write code that screwed up basic math but LLMs do it all the time.

[1] Obviously I am easily amused :-). But it is sort of the 'pelican riding a bike' test where you can easily tell what a human would do vs what an LLM does.

[2] EDIT: another commentator says they vibe coded it in a weekend. That would support the hypothesis that the application could be at lease a small number of shots vibe coded. That said, I'd love to see that code and the prompts that led to it.