It's utter garbage. I wish they'd design the game concepts with the idea firmly in mind that it should be something which is easy for an AI to cope with.
This, definitely. And I think the players themselves deserve blame here not just Firaxis. I think the playerbase as a whole dogs Firaxis to have this and that cool feature, civs, graphics issues, etc. that those things get made the priority, because they're what we playerbase buys the game over, when they're just building something the AI is never going to be able to figure out how to compete in half-decently, or investing resources in those things rather than in developing the competitive side of the game. So for a quick sexy two-minute rush of cool we demand the developer sacrifice the core game experience.
At risk of beating a dead horse, I remember reading a lot of talk of 1UPT, main points being that it felt cooler, funner, looked better, etc. And sure, that's very true. Exciting 'tactical combat' - as if you could call tactical the equivalent of playing chess with your opponent not knowing how to move any of his backline pieces. In a better world, with a supercomputer, we could make it work. But in the real world, with budget constraints for the devs and processor constraints for the players, why is it a sin to cut our losses and accept some combat mechanic that the AI can handle better (not perfectly) at the expense of some of the sexiness? AI can be taught to put together a balanced stack, and make semidecent decisions regarding when to advance it, when to retreat it, etc. It is much harder to force him to split out those pieces and move them all optimally, concentrate force on the right points, etc. There's twenty variables rather than just 2 or 4. Things the human does naturally. AI is much better with vertical than with horizontal combat (can kind of see this with ranged units in Civ 5 at least, the AI is semidecent finally at getting them to focus-fire on one target)
This goes for other mechanics too, and no civ game has been completely immune to the problem. But it certainly needs to be minimized as much as possible - the less we need to kick up the artificial difficulty hurdles to make it a challenge, the better.