I think I may have misunderstood with the mana pools. If you just mean mana pools as an internal mana budgeting scheme, that's probably a good idea.
There is a big difference to me between an AI with bonuses and an AI that doesn't play the same game. Here is one economical (in development terms) way to make an AI. First, forget about managing the economy. All you are interested in is the appearance to the player, so just spawn units at the rate and quality that seems to match the difficulty level. You put buildings in the cities along a timeline consistent with playerlike development, so when the player takes the cities it looks like a city that was doing stuff. When the AI declares war, which it does at a pace that the developer thinks is consistent with fun and difficulty, the AI sends stacks directly toward your cities. No scouting or anything, because it is actually a tremendous amount of work to keep and work with a belief state in the AI consistent with the knowledge a player would have in the same situation. The AI expands at predetermined rates for the appearance of human-like play, but there is no resource consideration going on.
I think you get the idea. That kind of AI would save an incredible amount of devolopment, and give a decent illusion of a human-like opponent if done for that effect. And you can make it as challenging as you want it, not hard at all. But that is also not what makes writing game AI for a game this complex one of the most intellectually challenging activities that one could engage in. The AI I described isn't playing the game at all. Giving bonuses alone doesn't produce the situation above, the AI must still utilize them through playing the game.
Playing the game well for a human is extremely hard, making an algorithm of good play that uses the same rules as the human and has belief states consistent with human knowledge in the same position is a task at the edge of humanity's knowledge. I mean trying to make something like a Deep LH. I wrote a thread about that before, but if people took up LH as seriously as say chess, I don't think an AI could be developed that could come close to competing with humans, based on the relative complexities of chess or go versus a game like LH. I don't know quite how the numbers work out, but there may be as many starting board positions (with monster placements, wildlands, map varieties, different starting locations for the factions, ect) with the default sovs in all the combinations of starting variables than there are possible legal board positions in chess. That is the tip of the iceberg in complexity analysis, which I did in more detail here and in the replies https://forums.elementalgame.com/438446/page/1/#3300617 .
"In go, for instance, you can only do one thing in a move, play a stone on the board against a single opponent. But in FE there are way more complexities: multiple opponents, it is possible to make dozens of moves within a turn, far more complex rules, the stochastic elements (the gameboard, placement, and events are random), enormous variety of game "pieces" (spells, units, champions, monsters, resources ect), and the fact the number of things that can happen and the number of things you can do in a turn increases as the game goes on."
In other words, you can never use minimax on LH.
