Other than select software being pissy about it, not much. Just like how there's nothing wrong with CRLF, except for select software being pissy about that too.
My problem is that what I input (and observe!) doesn't match what's persisted. Worse still, editors lie about it to me until I close the file and reopen it. And just to really turn the knife, various programs will then throw a fit that a character that I did not input and my editor lies about not being present, is present. I hope it's appreciable why I find this frustrating.
I expect my editor to do what I say, not secretly(!) guess what I might have wanted, or will potentially want sometime in the future. Having to insert a newline while concatenating files is a chore, but a predictable annoyance. Having to hunt for mystery bytes, maybe less so.
"Unix programs" I haven't ran into throwing a fit per se. That's why I didn't write that.
What I ran into issues with was contemporary software that's shipped to Linux, such as Neo4j, which expects its license files to have no newline at the end of the file, and will actively refuse to start otherwise.
I have a feeling I'll now experience the "well that's that software's problem then" part of this debate. Just like how software not being able to handle CRLF / CR-only / LF-only, is always the problem - instead of text files being a joke, and platforms assuming things about them being the problem.
I know it's just me but my worldview is that the world would be better if all editors had "insert final newline" behavior