There's some good AI out there though, for turn-based games particularly, but there's time to run an algorithm based on events _not_ happening in real-time, so it's less daunting of a task vs. a real-time game.
If you compare the AI from the published CC games with the last ten years published RTS games, with a few exceptions, you will see a lot better AI.
Anyway, comparation here is a bad idea because the published CC games used a game engine from more of 22 years ago where deep changes in the AI were not possible and players usually think as if they were new games where they were not.
For me, I would be happy if the AI attacks me and it uses some type of tactics. I do not wait miracles, only some from the year 2018, not from 1996.
I always wondered why developers didn't just apply textbook tactical doctrines as responses to situations. Add some variables (morale, insubordination, FoW, communications breakdown, timing, etc.) and that would be a quality AI.
For example, taking out a fixed position: suppressing fire, flanking, smoke, speed, these are all things that shouldn't be that difficult to quantify.
We're not asking for Napoleon in our AI, reshaping the very concept of operational warfare, just textbook tactics implemented on a script when certain conditions are met.