After playing a couple of FH games (including on fairly lengthy one, all on challenging world difficulty and ai difficulty) i think i can confidently give the following feedback:

Ai:

In all my games the ai was decent enough in the early game, falling a bit behind when it came to the mid game and was outmatched (by at least one order of magnitude (by point rating in the upper right corner of the screen) in the late game). My assessment is that the Ai has no map awareness at all.

Here is what i do: Use the sovereign with the least amount of support troops necessary (else they steal exp) to clear an area that has good tile yields from monsters, especially the lairs. Rush production and unrest reducing buildings in the new founded city and give it unrest reducing enchantments or production increasing enchantments if possible.

When i look at the Ai empire and kingdoms there is one glaring difference: it does not clear out monster lairs as it should, this leads to the issue that the ai looses cities to monsters even in the very late game. I am not talking about an occasionally loss, it happens quite often and very predictably.  Usually ogre lairs are the most common threat to ai cities because the armies that they spawn are very capable of defeating city garrisons and the lair defenders are usually strong enough so the ai will not touch them. While this is fine up the mid game, in the late game when the ai easily has the capability, it still does not clear out the lairs and constantly looses cities. This has dire consequences because only a very few ai cities are even close to an average player city.

The problem gets much worse if the ai settles next to wildlands, this will lead to a massive attrition because of monster armies leaving the wildlands, attacking cities.

On one hand this is an strategic Ai issue, on the other hand it is also due to horrendous outcomes of automatic battles, which is the only option for the ai, unless it fights against the player. If i had to rely on automatic battles only i would loose much more battles instead of the very very few when i misjudge the difficulty of a fight by a slight margin.

On all other accounts the ai does a decent enough job, although i have not looked deeper in it's city development for example. with the current game mechanics the following strategy is obvious:

1) taxes should always be low. there are very few reasons to ever raise them (this is a game design issue, currently you might as well remove tax levels)

2) city placement should be 2 food minimum (i saw the ai found cities on 1 food tiles), 3 production minimum and at least an essence if possible

3) city development should follow the order i described above: gain effective production (production increasing buildings+unrest reducing buildings) as fast as possible, i.e. rush them if possible.

4) research should follow that pattern and only add weapon and  magic techs if necessary. 

5) only produce combat units out of fortress cities which have all unit boosting buildings available (rush them too!). i often see the ai building inferior units across its cities, wasting production instead of building up its cities.

Balance:

The assassin is the strongest class by far and only because of one trait: +1 dodge per level. I easily get my sovereign and heroes to achieve 70+ evasion rating (in my last game my sovereign had 120+), which means that the only unit to ever harm (not kill, just harm) him would be a very high level hero with +1 accuracy/ per level. secondly the trait tree has two outstanding abilities: the one where you swap places with an enemy unit and get another turn and the guaranteed crit. those two abilities negate casters, archers or any key unit in an enemy army. secondly the armor ignoring traits lead to a much higher effective damage than a warrior with much higher base damage. In my opinion you either have to introduce normalized distribution to accuracy and evasion so you never gain 100% chance to hit and evade or just remove the +1 dodge/level trait. It is heavily abusable by the player.

On that note, lightning hammers should be changed as well. +1 lightning damage per level means that there are only very very few weapons that have higher damage. The issue was fixed for lightning pikes a long time ago, why was the hammer ignored, especially since it comes so early and is very cheap?

Another issue with evasion is that there is no real gear for, nor against it. meaning even in high armor gear you can achieve high evasion rating. i don't know if it is intended that assassins run around in full plate (champion armor), dodge every hit and wield big two handed weapons which ignore almost all armor. In my opinion there should be dedicated evasion gear and armor with high defense should reduce evasion. This difference between light and heavy armor would make a world of difference in strategy, thus replayability. Currently the only difference between light and heavy armor is the weight, i.e. you sacrifice a few initiative points in favor of a lot of defense. even if this was the only distinction between light and heavy armor, the fact remains that only early game armor and is light and all late game armor is heavy.

Now onto cities:

Certain buildings can lead to heavy abuse, most notably buildings that grant empire wide bonuses to other cities, such as +x food per grain for all cities and +5% hp for all units. The former can lead to 2 food tile cities being able to reach level 5, the latter inflated hp numbers for all units. there comes a certain empire size (i.e. number of cities) where those bonuses yield far too good results, i.e. you will never take the alternative when leveling up your cities.

Other empire wide bonuses such as reducing unrest empire wide, like the onyx throne, are pretty strong, but they do not lead such game changing outcome as the other ones.

Art:

Let me put this out upfront: i do not like the 3d art style at all. It is a major contributer why i do not like the elemental franchise in general. I do however like the 2d art, especially the cloth map, i only wish there as an option for tactical battles in cloth map mode.

Ignoring my opinion, i think there are way too few assets when it comes to different armor types. in every game my units look the same at some point (all champion armor) and there is little difference (if any at all) between kingdoms and empires in armor style, let alone factions. This is a major contributor to an overall impression that everything looks and feels the same and heavily contributes to visual boredom.

The same is true for cities, unless you zoom in and look at the detail, all cities look the same from a normal viewing distance (look at any city for 1-4 seconds and see if you find any outstanding visual objects). personally i would like an approach as in HoMM5 where cities look widely different between factions (and they are good looking too) or an art style as in cIV/ciV where cities may not have identifying visual qualities but the size is easily recognizable (i.e. you have a visual clue for your most important cities in your realm) and abstract enough to not break the immersion.

I understand that when it comes to visual diversity, a lot of it is dictated by memory usage. If the current art assets are the maximum memory wise, i highly doubt the visual diversity is distributed to match the gameplay distribution of art assets, i.e. you have a lot of tiny different things that rarely gets noticed (like unique looking boots that can only be looted) while the mot common things share too much art.