Looking only at the campaigns (and at Normal difficulty -- which probably doesn't effect the AI itself, just the resources avaialble), the strategic AI seems to be very conerservative in comparison to previous installements -- turtling is a valid strategy in most campaign scenarios, for example. Specific issues include:
- Hero army growth is far less than you would expect -- even when the enemies have 3 cities (to your one), you seem to be able to outgrow the enemy heroes.
- This may be because the AI sets a target for army size -- when that's reached, it sends it off to "adventure". Once you are able to beat that army with negligable casualties, attacks cease
- AI doesn't seem to build up their cities very agressively, which may be a partial reason for the previous problem.
- AI is bad about collecting loose resources -- it isn't unusual to see unclaimed, unprotected, resources as late as the end of M 2 in enemy held terratory.
- AI is very conservative about evaluating the outcome of fights -- this may be beause the tactical AI is... Bad.
Overall, I would say the strategic AI is passable, but not award wining -- given the scripted advantages in the campaigns, it achieves its goal of making you work for success.
The tactical AI, on the other hand, is awful by any defination:
- The AI never uses the wait command. You can, for example, always wait on your first turn to allow enemy melee to close the range so that you don't take the half damage penalty on your first shots at them.
- The AI never uses the flee combat command. Arguably, this is an improvement, as the AI fleeing is always very annoying...
- In castle defense situations, the AI is *very* aggressive about leaving the castle walls.
- Level 3 fortifications are far, far tougher than they have been in Heroes for awhile -- unless you have the skill that boosts siege weapons, it is a safe bet that you won't be able to get through the walls for 4-5 turns, perhaps much longer.
- The AI handles melee units adjacent to shooters *very* poorly -- 9 times out of 10, the shooters just defend. Tthey /should/ move away from the attacker, with a minimal goal of ensuring one melee attacker can't be adjacent to more than one shooter at a time.
- The AI doesn't have logic for "I'm going to lose this combat, might as well do as much damage as I can before I die" -- casting spells like stone skin on its last unit, rather than direct damage (if nothing else, the default hero attack) instead.
- The AI targeting priority is broken -- if the enemy has 1 stack of 500 skeletons on the far side of the board, and has summoned 4 water elementals adjacent to two stacks of shooters, the AI /always/ targets the elementals.
- Oddly enough, this "blind spot" is limited to the first, least powerful, unit (skeletons for Necro, Fairies for Sylvian, etc.) -- if you have a stack of 500 skeletons, 300 ghosts, and 4 water elementals, the AI will (more correctly) target the ghosts.
- The AI simply refuses to admit that the fire wall spell exists at all. Place it on several stacks of shooters for guranteed damage for the next three turns, place it on the map where you expect enemy units to move to (e.g. "max move towards enemy units") and they will reliably walk / fly / teleport into it.
Basically, the tactical AI is only a threat when it has an overwhelming force advantage to begin with (and, to its credit, the strategic AI only attacks when this condition is met -- if the AI attacks you, expect to take /heavy/ casulties). In the campaign, this is acceptable -- but in a skirmish map...
https://steamcommunity.com/app/321960/discussions/0/490121928356092029/#c490121928358212093
