nobody with brains installs a GUI on the server.
And will someone PLEASE tell IBM and Oracle that?
I like Systemd in concept. It potentially allows setting up dependencies from the outside, much as Inversion of Control does in software. Meaning that the systems don't need to know as much about other systems because it's wired into the overall system configuration. And, unlike init scripts, you can make the management of subsystems dependent on the actual state of other subsystems, not simply assume that because one was scheduled to come up before another that that's what actually happened.
However, the one thing you should absolutely positively NEVER do is replace a major product with one that lacks critical commonly-used functions of the original product, and that's the fatal problem with not only systemd, but Gnome 3, and some would argue various versions of Nautilus.
And when developers ignore the angry mobs, tell them that they're unappreciative or stupid, or otherwise incapable of recognizing a Superior Product when they see it doesn't do any good for anyone. If your new shiny toy doesn't cut it, you either need to add those critical functions before shoving the old system off the pier or admit that your design is too flawed to handle it and go back to the drawing board.
I like systemd in concept and am prepared to become a full convert. But only when it restores the essential functions that systemd provided.
However, I totally loathe systemd's partner in crime journalctl and frankly don't see myself ever learning to love it. Too much replacing simple functions with complex commands and too much opacity in the log storage itself.