True enough, it's a common problem in game design, in fact there was another thread here someone started discussing (as I recall) games that had effective harder settings.

The solution, and this is me speaking from no knowledge of how this works, would perhaps be to play back into an AI's strength by pre-setting tactics based on watching hundreds or even thousands of human vs human matches (not like CA couldn't collect the data). The patterns of engagement aren't infinite, or even close to it, that would then be the tactics you would program to, making it a series of steps the AI follows. And of course it would be even better if they could figure out a way to tie back results to modifications in subsequent battles the AI fought, course I think that would be rather hard. So the brute force approach would be the first place to start.