The company I worked at maintained some ETL pipeline on behalf of Yahoo, and the strict guidelines from the client and the ones developed internally stayed you from using anything coming from CPAN.
That means that you didn't get to have function signatures, as those were a package back then, you had to declare all your arguments at the beginning of the subroutine body by taking them off the implicit parameter variables.
Programming in it wasn't crazy for sure, but it just felt like a really clunky Python. The split between the data structures that could be handled by reference and the ones that didn't was very unfortunate, Yahoo mandated that you only use the ref-type, and you had to be converting back and forth.
Well, that was part of my experience working for some 6 months at a team where strict discipline kept a big system working, but there was little upside to using Perl. The worst was the error handling in this scheme, keeping your own makeshift stack trace with backing up the error variables was a hassle.
This is what kills it for me more than anything else.
Perl's syntax is incredibly unintuitive. People joke about it being the only "write-only" language because you can write some code and 6 months later come back to it and have no idea how any of it works.
Code is read far more often than it is written, so readability is important, and Perl lacks it.