Indeed, in previous page (i think), i did say there are some factors that influence why the AI is as it is (without any mod). Which i found rather weird.
There is no mod trying to fix it so far because it's not that simple, there are other factors, and i suspect hidden factors which modding tools cannot access.
I'll list what i found so far.
The AI has different definition of game phases (as said by those posters in steam forums), late game is turn 55, i was told that some people said there is phase(s) later than late game, but i suspect they have either mod it themselves and forgot that they did or download a mod which include new phases there. Because my purely unmodded planetfall so far only has 3 definition of phases for the AI. Start game, at turn 1 - 30, by 30 the game define it as mid game, turn 55 is defined as late game.
However, there are other more complex factors involved in AI decision as to why the AI doesn't mod their tier 3 and 4 units, and why they produce them in a so small number relative to time passes, and relative to other players the AI consider as threats. First of this factor is unit priorities, by default their unit priorities regarding tier of the units are set to be the same from tier 1, 2, 3, 4. Then other factor in their decisions is heavily influenced by other players's military capabilities the AI deems as threat, this players include other AIs, not just human player. It is often the player the AI met first, this is why if you team the AI, they become dumber, because they even consider some NPC as threat (including NPC they are actually at peace or friendly terms with), where NPC faction military power is stagnant. FFA AIs, make the AI better, i've already theorize this long ago (at vanilla planetfall), before i even check the modding tools, now after delving deeper, it confirm my suspicion.
The AI will only produce units, in repeating cycle of a set, each set is determined by a value of combat power limit set (fortunately this value can be modified in modding tools), until each of their stack has a combat power relative to those factions's best stack (or overall best stack) the AI deems as threat, this also confirm my long ago theory that is this is the reason why AI stacks even at turn 100+ only stays at 1k-1,5k combat power. I've seen 2k+ long ago but rare and only turn 80+ that these rare event happens.
There is also other constraint factor, that is the resource the AI is allowed to allocate to produce those units. By default it's very small, no wonder that is why when the AI produce T1 and T2 units they are modded, but when it is T3 and T4, they are totally unmodded (fortunately this limit can be modified). But because the modding tools doesn't allow to peek deeper let alone change further than this, it's making it quite hard for modders to see and understand how all the important variables that influence AI decision beyond what the modding tools allows affect each other and ultimately affect AI's decisions.
Other notable factor is victory types heavily influence AI plans, from my experiences, most games often set a general base plan for their AIs, then modify this and there for specific goals for specific victory conditions, that is not the case here, at least from what i've found in the modding tools. To illustrate what i mean, for example, if you disable doomsday and enable domination, then the AI plan for the murderous type of goal is active, but they will so neglect their research thus actually unable to murder anyone, so to provide the best AI, players need to enable all victory conditions. In my unmodded games, AI often race for doomsday victory at turn 50-70 (i never disable any victory conditions), these always happens before i checked the modding tools, now i understand why.
Regarding AI decisions on upgrading their units's mods, they have priorities too, i haven't check them deeper though.
The tools also allow to dictate the AI in a quite step by step way, especially strategically, like setting a complex plan for them, this will require time to experiment for us modders to actually create good result regarding improving AI performances, because these resources cannot be peeked and delved deeper too, so it's going to be quite tricky endeavour to combine quite a simple set of instructions to make a complex bigger instruction.
I did an experiment to see how far the AI can be pushed, sometimes, the AI do succeed in what i told them too, but sometimes they stray in a very unexpected results. When they do succeed, they manage to produce 5-7 bombardons, 7-11 harriers, several other T2 and T1 at around turn 20-25, mods wise they use a mostly T1 mods with a little bit of T2 (which is intended as i haven't modified any resources related to AI unit mods), not counting those units they purchase from NPC, but disabling NPCs factions make this result more constant, which is why i'll test it without any NPC for now. And this is the interesting thing, their primary priority is not capturing colonies nor sectors of human players, but to hunt down the human's players best stacks, this result in the human player cannot defend their colonies and forced to be nomads, then i try making sector garrison nonexistent (which also affect the AI, but when they are too superior in everything and leave no hole in their territory unchecked, no garrison is no big deal for the AI), it's even very torturing for the human player, as it's a free territory for the AI. That is just one AI, imagine surrounded by multiple of these psychos. When they stray, they will just mass T1, and attack any hostile stacks without purpose. There is some hidden variables that influence their decisions here, which unfortunately the modding tools can't access, this is when it get tricky to make the result more constant whilst the modding tools only allow not so deep modification towards the value and the instructions are too few imo. And i'll tone down those psychos of course after i figure out how to make it more constant, this is just to see how far if they are pushed, if i tone them now, it will take me more turns to see and test them (even with console command a.k.a cheats), while i haven't much time nowadays.