The problem if cheating AIs is that it's impossible to play strategically against it. It's difficult to cut the AI's supply lines, sabotage its production or manipulate it if its power comes from arbitrary reasons. It's difficult to analyze its moves and progression if they don't obey to similar rules than the ones humans are subjected to.

If the AI cheats then how can the game be fun?  Everything you do and everything the AI does is meaningless and boring. The game becomes meaningless and boring.

In 0.912 I remember often watching a AI army prancing around acting hard, then getting kicked on its arse by some wandering beasties.  Not only did this make me laugh but also made the game feel real somehow.

I don't think this has anything to do with AI competitiveness, it's all about how you like to play the game. Basically, it's the difference between quick domination versus slow, drawn out war. You might like to play quick simple games, and that's fine, but not everyone feels the same. Heck, sometimes, I feel like the movement is still a bit too fast (particularly because of roads) because as soon as you break the front lines the war might as well be over, since there's just no time to muster a feasible defense to counter attack.

Also, when you move too far, too fast, the world feels small, conversely, when you slow movement down, the world naturally feels bigger.

I'm with those in favor of AI competitiveness, but you do realize that the AI does cheat at higher (challenging and above at the very least) difficulties?  At Challenging they all get a gold mine (though I've never seen one put a mine on it, even 270 turns in), at Hard it's sometimes 2 gold mines.  They can cast spells like Tireless March at turn 1, and can double cast things like Inspiration and Obsession (even when they shouldn't be able to cast Obsession at all).

I don't have a problem with these things, and there's nothing wrong with cheating AI at harder difficulty levels, but it needs to be done intelligently.  And, for the most part, it is being done intelligently, so good work Stardock. 

I concur. 

One main change I'd like to see is the exp for heroes to be based on map size.

I think the current pacing is geared toward the hardcore player and really should not be the default setting. Casual players will want a faster pace game where their champions feel powerful. The current pacing works when you have a good understanding of how the game works. One thing I like about the current pacing is that the game does not drag out unnecessarily. While the game did take longer to play, I did not have to spend half the night mopping up stray cities which I've found to be a problem in some of the older games.

I don't believe they are mutually exclusive goals. And in some ways they are complementary in that having a competitive AI enhances and makes the gameplay better (in the same way that playing solitaire chess isn't nearly as much fun as playing against a competitive opponent).

However I do agree that achieving both goals is much harder than just achieving one of them and in some places Stardock have deliberately taken the easy way out by removing things that the player finds fun in the name of making the AI easier to code. I would like to see less of this, or at least countering any such moves by adding back other things that the player finds fun.

In my original post I used language that expressed intent instead of effect. I don't know what the intent was behind the changes made from earlier betas, but the effect is a game that is a more manageable AI problem. But the thing is it will never be enough. Games like this are problems that are far harder than Chess and Go from an AI standpoint. To list some factors that make up this huge gulf of complexity: there is a larger number of both candidate moves and actual moves per turn, there is the fact there are multiple opponents, the rules are made up for this game and are not well-studied problems, and there is the randomness of the world as a hostile actor. There is simply no way anyone can write a straight up AI (anytime soon) that will beat the hardcore players on here, not even give them a good game. I think this is realistic. Now if the rules are tweaked to play to the AI strengths which are actually the human player's weaknesses, then maybe it could beat most people who play the game, but still nowhere near the above mentioned class of players. The sacrifice to do this would be that the human weaknesses are a lack of attention span over long periods of time, a distaste with micromanagement, and they want the long-term goals to arrive sooner so they press the turn button to get there faster instead of maximizing their moves. The result of using this against them would be at least some players (I suspect many) will have less fun.

I think the straight up AI is a good thing and it should be a setting in the game, but I don't think any of the hardcore players should expect that it could really challenge them (barring advantages like being immune to dealing with a hostile world). Given this, I think the focus on challenging players should be giving the AI advantages that make them mimic a human's effectiveness, while leaving the human the most exciting game possible to play.

The World (monster) AI should be challenging.

The Faction AI should try to win via one of the many victory conditions.