Kind of weak. It's possible to declare war on a neighboring civ and just waltz your sovereign into their city. Also, enemy AI like to leave their sovereigns exposed. I wiped out 3 of the 5 civs I played against in about 10 minutes just whacking their unprotected sov's.

its alot better in 1.05, but still has room for improvement. if you rush, you can wipe the ai out pretty easy, but seriously, what fun is that? it gets better in the late game. but like i said, still room for improvement.

I rate it very highly. I'm playing on a small map with 5 computer ai opponents with extreme difficulty and getting nice interaction and challenges from them. Next game I'm going to put maximum on the map. I remember in Master of Magic the best challenging games were on small maps with max ai opponents which was just 4 back then.

Incredibly poor.

Even on large maps, "ridiculous" setting and just sitting back for a couple hundred turns before I make any aggressive moves towards the computer players, it's a pushover.  Most of the AI controlled cities will be very lightly defended (if at all) and my only real danger is if:

(a) I'm relying on a single killer stack

(2) I'm too far away from a threatened city and too far into my own territory to cast teleport.

Even then, it's just one city lost.  The AI rarely pursues that sort of conquest aggressively and is quite satisfied just sacking a single town.

It's interesting reading posts that say the AI is poor.  My experiance is that the AI is very conquest heavy.  If I don't research the conquest tree (and currently (for me at least) the conquest victory option can not be turned off), then the AI will steam roll me.  It's even worse in the early part of the game, if the AI happens to be near by and I've yet to build up a army.  Once they declare war, they'll throw wave after wave at you without a break, meanwhile you have to balance your troops/econ to fend the AI off.  It has also has a tendency of slipping past the town which has your main army in it and going for other towns which are not so heavily defend, which means you have to have troops garrasioned in your towns or risk losing them.  The AI also seems to like spamming pioneers and establishing undefended outposts a lot.  Anyway that's been my experiances with the AI.

Its ok. I haven't tried the imbue thing but it sounds like cheese to me.  Exploiting a completely unbalanced aspect of the game then saying the AI is no good is rather silly.

I think the AI in starcraft 2 is decent but it can't handle things like a protoss cannon rush. Protoss cannon rush is cheesy just like the imbue thing.

It's a waste of time to build up the AI to combat that strategy if the magic system is changing so drastically. I'm not really expecting a whole lot out of the AI until the game mechanics settle down. Trying to hit a moving target with AI strategy.

They have made some simple improvements that are making less suicidal with their sovereigns and that should help no matter what the mechanics. I appreciate that.

Archers, deflect, and that ice stabby spell are enough to kill anything.

Pretty much this although I haven't played through an entire game, due to the ridiculously easy AI, and so not sure on how long it takes.  Once I start easily conquering my neighbors, the game becomes too boring and I quit.

The AI is very poor.

To somewhat compensate for this, the game is so badly balanced that if you really want to you may be able to convince yourself that it is all down to game balance. If self delusion is your sort of thing, that is.

AI also does not use spells. I've never seen it use teleport when it should, and gets stacks of units "stuck" constantly between borders, influence zones, terrain features, etc. 

I have a pretty lengthy list of things I want to play around with this month and next (for starters).

If you guys want to see what I've been doing in my spare time, check out this video:

Now, in case anyone wants to know why AIs are generally done this way the reason is that it's commercially unviable to build AIs that do lengthy-multiturn planning because you can make an AI that challenges players without having to do this (either through cheats or because players just generally aren't that good).

But multi-turn planning allows for a much more intelligent AI. It just takes a LOOT of work and for the developer to be comfortable managing a lot of concurrent threads/cores which is something I particularly enjoy doing.  But if you want to get an idea of what I'm going to be doing on my extended time off, this is the bulk of it.

The AI is going to keep getting smarter, till it is unbeatable (Its inevitable, like Gary Kasparov eventually being defeated by Deep Blue + AI  ). In SoaSE, I got good tips from watching the AI in recorded games (eg build a Capital Ship factory *first*).

On a setting of "ridiculous" the AI should *always* beat the human regardless of percentile unless the human gets *extremely* lucky. That's what you should shoot for on ridiculous. Perhaps "hard" only the 1% can beat consistantly, perhaps "medium" only 5% beat consistently all the way down to "easy" in which players on their first game have a 50/50 chance.

That's what I would shoot for anyway.