1. Power rating effect on diplomacy. Absolute difference between power ratings is a poor basis for AI decisions, since starting a war with someone at power 200 when you're at power 300 is an entirely different matter than starting a war with someone at power 900 when you're at power 1000, and both are completely different from fighting a war with a power 100 enemy when you're at power 200. Relative power ratios would seem more appropriate, as if I only have a 10% power margin over one AI and there are other potential enemies present, I'm not all that likely to start a major war, but the AI would if the absolute difference in the power ratings is greater than its 'you're much weaker than me' threshold. It would also be nice if there were a way for it to understand when I'm only fighting a limited war (clearing the AI's units out of an area near my territory), as opposed to launching a major assault on its continued existence, but I doubt that that is going to happen.
2. More diplomacy stuff: if there's one AI at 900 and another at 600 and I'm at 400, I should be a tempting target for the 900 power AI, but the 600 power AI should think twice before going to war with me, and should consider me a good candidate for an ally against the 900 power AI, since together the 600 power AI and I are theoretically more powerful than the 900 power AI, especially since a war with me while the 900 power AI is in the neighborhood is likely to weaken us both to the point where the 900 power AI could come in to pick up the pieces. In other words, make the AI consider a 'big picture' for the world (though limiting it to things it should know about is reasonable - if the 600 power AI in my example doesn't know about the 900 power AI, it isn't too bad an idea for it to try to invade me if I'm at 400 power, unless it has other problems).
3. Even more diplomacy stuff: we can see relationship modifiers that the AIs have towards us. Can we either:
See the relationship modifiers that the AI thinks we have towards them so that we can see the modifiers that the AIs think we have towards them;
Or, have the player able to set relationship modifiers towards specific AIs (say, a slider bar ranging from 'Friendly - we should become allies' to 'Neutral - I may not be fond of you, but I don't think I'll go to war with you without great provocation' to 'Wary - don't violate my territory or I may declare war' to 'Hostile - I'd as soon kill you as speak to you' kind of thing) to try to give the AIs a clue as to how I'm likely to treat them (which should be limited in abuse likelihood by some sort of rating the game gives me for my warmongering tendencies or whatever I've been doing that game). It would also be useful for the AI in deciding if it should consider alliances or other treaties with me, and how it should treat me - if I say to Resoln 'I love you guys - we should have a non-aggression pact or alliance' but tell Resoln's ally Yithril 'I hate you guys - I'd declare war on you as soon as look at you' then unless Resoln doesn't like being Yithril's ally it should probably be wary of making deals with me. This, of course, assumes that players would set their diplomatic leanings honestly - but if they don't you could have a 'treacherous' modifier (or set of modifiers, depending on how often diplomatic dishonesty occurs and how serious the treacherous action is - say, declaring unprovoked war from 'neutral' or 'wary' is less of an offense than from 'friendly') that makes AIs much less likely to look kindly upon the player.
If you go with showing the player what modifiers the AI thinks the player has towards them, can we have some hints in the game manual for how to affect the modifiers the AI sees? E.g., if the player has picked up the 'I'm crazy' modifier, have some sort of suggestion (either in a tooltip or in the manual) on how the player got this modifier (by being a nutcase who offered an alliance and a gift, then declared war, and now wants a nonaggression pact but demanded 800 horses with red saddles), how to strengthen it (perhaps continue to be unpredictable in your actions towards AIs, giving them gifts one turn and declaring war the next), and how to get rid of it (by acting more predictably or in a more understandable fashion).
4. Individual unit movement-point tracking and the other movement point revisions suggested earlier in the thread.
5. Tactical combat unit positioning. It's gotten better, but there are times when the autopositioning of units really isn't how I'd want it to be. If the maps were expanded so that there was always a space of one or two turns to arrange troops before melee units close the range regardless of how far up the tech tree the players have climbed, or if there were a manual placement phase, or if autopositioning were simply better, it would be nice. I don't mind ranged units and spellcasters being able to attack on the first round of combat, but having relatively normal melee units able to attack on the first round on most tactical maps gets a little annoying and lessens the value of ranged attacks.