Here's what I've got to say.

1. The AI should probably harvest crystal and metal faster on higher difficulties.  They unlock the tech to use it for things VERY fast, but aren't able to make strong armored units because they're still harvesting material very slowly.  Yithril gets a big advantage for a lot of the game simply by virtue of having access to juggernauts, which, of course, do not require metal.  Eventually Gilden will pull ahead, but it takes a long time.

2. Cities need more transparency about how they tick.  I'd really like to see not just the turn cost but also the resource cost of buildings; some buildings take more than 1 turn to make even when they say they will only take 1 turn (elemental shrines and temples).  I REALLY would like to see a more explicit accounting of the bonuses cities get on level up; I'd been playing this whole time not even realizing that when a conclave leveled, to three it got extra research!  I had to mouse over the city’s resource breakdown to find out, and then compare it with a lower-level conclave.  

3. Champions kind of suck.  Commanders can be BANNANAS, especially in the right army (or against Resoln); tactician and trainer can get kind of nuts.  Mages can be great if well-built.  Defenders are decidedly ehhh, but ok for the stun, SR and shieldwall.  (I'd really like to see flat defense bonuses as opposed to "extra defense when defending".  Why are you giving me neat abilities if you don't want me to use them?  At least cap that tree with the racial ability that lets me make a guarded strike.  Or SOMETHING.)  Warriors and Assasins are TERRIBLE.  They underperform trained units with alarming consistency.  I almost never get worried if I see an AI champion.  A lot has already been written about champs and leveling; I'd echo the recommendation that the leveling ought to be faster, and better.  But I'd also like to point out that Assasins and Warriors have a lot of overlap: fundamentally they both deal damage.  Maybe it would be a good idea to merge those two trees.  Alternatively, the commander does two things: administrate cities and lead armies, which are radically different.  Perhaps that tree should be split.

4. The AI is so bad at tactical combat.  One area that could really use improvement would be target selection.  If the AI melee units can move to a target this turn that already has an attacker on it, they should do so in order to swarm.  Instead the AI just keep hitting its already engaged target, even if it's ally is engaged with an enemy it could easily swarm just a square's move away.  Battle for Wesnoth is an open-source tactical game that has entirely solved this problem; may I suggest taking a look at their code?

5. Archers could use some love.