Yeah, the differences between how single-player AI works vs balanced multiplayer AI is night and day.
For example, if you play a game like CKII what are the basic decisions you as a human take at the start? You might look around at weak nations and decide to absorb them into your growing empire. However, if every computer-controlled player also acted like a human player, you'd nearly instantly see every minor faction destroyed by their nearby larger neighbors. And this would raise some serious narrative questions. If consolidation started instantly as soon as the game starts, then how was the original map even stable to begin with? This breaks the sense of historical "slice of life" immersion. After that, the game would devolve into a WWI-like experience with a small bunch of mega-nations battling it out in a war of attrition, which wouldn't in fact be all that much fun, or very historical-feeling.
Also, one problem with having set scenarios where AIs make rational decisions, is that given the same input data, they should make the same decision every time.
Because that's what rational decision-making is all about. But no, a game of CKII where each faction always made the same exact moves (based on assessing the situation with rational logic) would in fact be boring as hell. So, instead of that, each faction needs to effectively choose what to do at random. Once they decide what to do, they follow a script where they mimic a human acting out that decision. But, so that the game always unfolds differently, whenever they are faced with a choice, such as which city to attack, who to declare war on, which other family to marry into, they're probably using simulated coin-flips to make each final decision.
So, no, they are not rational agents planning out how to win, they are randomly coin-flipping what to do, then following some human-written scripts on how to do the chosen action in a way that fools you the human into thinking the AI made a "decision" to do something.