I'm currently playing at Heroes of Might & Magic 5 : Tribes of the East, version 3.1 ... but I just can't wait for StarDock's first foray into the TB-fantasy-strategy genre. It's my favourite type of game. The Beta will be my biggest gaming event of 2009.
I don't mind if its visual quality doesn't match the level of HM&M 5 or King's Bounty : The Legend. StarDock is all about top-quality A.I.
The crucial problem with HM&M 5, even in its recent version 3.1 (August 2008), is that the single-player experience is diminished by the stupid A.I. of the computer players -- on all difficulty levels. I still play, once & awhile, because I enjoy the gorgeous 3D graphics, but the comps lack any strategic depth.
For example. The HM&M 5 programmers have given the A.I. a very simple instruction to follow : if the path is clear towards a weak enemy hero, or an enemy's capital, the top priority of its closest-to-the-tempting-target army is to rush towards the target. The problem, then, is that the A.I. most often does not take into account that a much stronger army can intercept it on the projected path. That's a frequent situation that a Human player can exploit.
It happened to me yesterday. Green (Elf) « saw » that the path was clear to enter my core region, where I had a very weak hero (a "mule") & my undefended capital's castle. It moved on the road towards that region ... but by doing so, it entered within striking distance of my very powerful principal hero -- which could crush it on the very next turn, a situation that was very obvious !
In terms of programming code, it must be extremely easy to instruct the A.I. to evaluate if taking a certain path will bring its army into the strike-range of a stronger enemy -- even though the target region is weaker.
The HM&M 5 A.I. is so obsessed by the priority of attacking weak heroes & weak capitals that it ignores the most elementary parameter of self-preservation. That is very sloppy programming by the Nival Interactive (Russian!) A.I. designers & coders.
There is also the issue that a castle might be empty of defenders ... but have a strong reserve of not-yet-hired troops. It can then be used as bait by the Human player : when the A.I. hero is near enough, you hire a hero & an army, and pounce of the approaching enemy. It's a well known trick.
The A.I. should then be allowed to "cheat" in this manner : its evaluation of the strength of the target castle should take into account its reserve of troops that can be hired and the gold reserve of the player who could hire those troops in time. The A.I. would then be allowed to assume that those troops are in a state of virtual existence, and it would not rush towards the tempting castle in those circumstances. Such an evaluation should be very easy to code ... and it would save a lot of A.I. heroes from that classic bait-and-switch trick.
I just can't wait for the 2009 Beta !