One person does not make a consensus, but here are a few things off the top of my head:
- AI is bad at naval invasions (both defence and offence) onto large islands that are not connected by land. Eg: British Isles is bad, Malta is fine.
- AI is better at reacting to attrition management than the player (because it notices instantly), but it's worse at planning for it (eg. knowing that supply will go down to 1k in the winter, so you either get a Winter Soldier or don't go there).
- The AI seems to use a completely different strategy for marriages than the player. Players try to use strategic marriages to forge alliances with people who can actually help them in a war (independent or vassal of same liege), or they breed for good traits/bloodlines, or to put their dynasty on a foreign throne, or whatever. The AI can sometimes do all of those things, but only by chance - it's very much a scattergun approach. Relating this back to your question: If you want to have epic marriage alliances, you'll need to script them.
- Following on from the previous point, everything the AI does is weighted random. IE: It very rarely makes the best possible choice, whereas an experienced player will probably always make a good choice. If you increase the power level too much (too many large stacking buffs) then the randomness of the AI's choices may lead to some crazy results. On the other hand, you could make it so only players are likely to stack those buffs (eg. going on a tour of warrior lodges for 5-6 special commander traits), which might give players a fun feeling of being powerful.
Thanks for your answers, this is already very informative!The AI isn't good at the game; there's certainly mechanics it is worse at dealing with (e.g. if it can't blob like crazy, it'll easily enter a Decadence death spiral it never recovers from), but it isn't particularly good at anything (e.g. it'll build buildings when it feels it can afford it (better than doing it when it thinks it can't), but it has no plan and simply picks something using a weighted ai_creation_factor for each building and whatever goes into its budgeting (not necessarily good)). It is also often not given what it needs to play rationally; there's a large amount of events that are just coin flips/die tosses without any AI weights (or with very limited weights), which can make a supposedly smart and sane AI character act completely against both its best interests and its personality if it gets such an event.
I feel it'd be easier to answer more specific questions regarding specific mechanics and how feasible (and time-consuming) it'd be to make the AI play better there than to try to answer it for the whole game, particularly as playstyles and personal preferences will affect what is "Good AI gameplay" (e.g. whether you feel the AI should play a character first or should try to provide a challenge for the player first, whether you feel it should be allowed to cheat with gameplay mechanics because it can't be made to understand them, and whether you feel certain things are fun to deal with).
Knowing what the AI is particularly bad at is certainly useful. If it can't really handle naval invasions well, for example, then I'll probably avoid having large islands. I guess I also just had some very frustrating experiences trying to mod EU4, where I realized that the AI literally doesn't use most of the new features introduced in the last couple of years, which gives players even more of an advantage than they already have. I'm not a big fan of including all kinds of features and playstyles just for the sake of variety, especially if they don't really tie into the core mechanics and if the AI can't properly interact with them; I'd much rather focus on a few robust systems and build something narrow but in-depth, instead of wide and shallow.
If I had to narrow it down when it comes to CK2, one of the first things that comes to mind are the different government types. I don't have much experience playing non-feudal characters, but know that they have some unique mechanics (manpower for nomads, patricians for republics etc). If I knew that the AI doesn't do well with playing either as or against non-feudal characters, I would probably not bother creating an equivalent for my mod.