Like a few other people, I still find the hardest AI opponents challenging. I don't have the time to learn to beat it.
But I look at the AI from a different angle: I've written AI.
Yup, actual pieces of software to play games (no, none you've heard of). It's not an easy problem, especially as the complexity of the game grows. Dom3's AI doesn't do so badly, actually. While the tactical AI needs spell blacklists (or at least player-selectable hint lists), it's generally pretty clever. The pretender-building code could be improved, tho.
But anybody can eventually beat game "AI" given enough time. Game designers are limited by two things: the processing power of home computers and the lack of machine learning. Your computer has one, maybe up to four, processors in it. Deep Blue, the big chess engine that beat Kasparov, had 32 and 256 special-purpose chess-playing chips.
Good machine learning systems are a pain to write. They also run very slowly. I've seen them run for hours on 16-processor machines while learning how to run factories. While it would be cool if Dom3 could learn to beat humans, I don't think that's gonna happen any time soon.