why arent they talking at all about the AI?

[Tellos Athenaios]

With designing AI there are always a few problems:

Who are you designing it for?
How much does it take for a slow system to make use of it? People with slow systems should be able to play your game by simply tuning the graphics down - and not the AI.
(Ever tried to create a snake game on a TI84plus with some AI to make the food move in order to escape from you? If yes, you know what I mean, even a relatively simple AI (with the above mentioned system and program this means: advoid obstacles, and if your route is obstructed try and flee the second best way away from the snake) slows down a lot.)
Do you really want it to be ALL seeing, and ALL knowing (this isn't realistic, since on a true battlefield most commanders wouldn't have been able to see ALL of it...)? If yes, how do you get it to respond in a sensible way?
How long do you have to develop it? I mean, an ALL seeing AI would require a self learning program - that's what they are into at universities and this has lead to 'speaking' systems. For companies however it would be nice if a new version of a series didn't take several years to complete...
How, when, and where should AI change it's battle formations? You can't just change your battle formations at any time, you can't do that at any place, and the key to do it succesfully is to do it in the right order, and in the best way adapted to the new given situation. To humans with a little sense of tactics this all comes natural but how do you copy the human way to the AI?
And to refer to my prior point of slowing down your system: to make effective use of battle formations and if needed to change those, the AI has to constantly check it's troop positions, movement directions, speeds, enemy troops and so on. And not only checking those, it has to adapt to new situations too, wich often requires entire new formations and always requires a new forecast. And so, a real killer AI in battle would also be one that takes processor minimal requirements to a whole new level.
And probably loads of other problems, to make it even worse. Such as the game is based on an earlier game wich had it's specific faults and bugs, and since this is some way just a new version of older software many of these faults and bugs can't be edited out without having to completely rewrite either very large portions or the complete game. Rewriting would then lead to a new game that simply doesn't fit as successor in the TW series nor as a start of a whole new series. (Often rewriting such large amounts of code requires rewriting other code not concerned with AI too - to make sure the comletely changed AI is going to work you might need a new 'enviroment' as well.)

Now why don't we see improved AI (yet)?

This is, in my opinion, simply a matter of REALLY good AI can't be seen (as it would seem as if as human was in charge and not the AI), improved AI is hard to notice (since this is mostly noticed in detail or by trying it - but there is no demo yet so this way in cancelled out for now) and to show AI you need to show a lot of your game. And all we have seen for now is just to build expectations not to give us answers on these sorts of questions: a company probably doesn't like when it's new products lose all of their 'surprise' values. So we don't get to see everything that's in the new game - which I feel confident about will include an improved AI. I'm very positive about this, since I doubt CA wouldn't make use from some of the excellent work Modders have done on AI battle formations like DarthVader's formations mod.