Computer opponents in strategy games are widely described as either predictable or unfairly advantaged. Both descriptions are usually accurate, and they follow from the same underlying difficulty.
The decision space is enormous
At any moment a strategy game offers a very large number of legal actions, and each leads to a position with just as many. The tree of possibilities expands beyond examination.
Games with simultaneous action compound this, because there is no discrete turn at which to evaluate and choose.
Search techniques that work in board games therefore do not transfer directly, and the opponent must rely on rules and heuristics instead.
Heuristics produce recognisable behaviour
Rule-based opponents follow scripted priorities: build this, expand at this point, attack when a threshold is reached. The behaviour is competent and consistent.
Consistency is precisely the weakness. A player who observes the pattern once can plan against it indefinitely, because it does not adapt.
This is why campaign opponents feel challenging initially and trivial later. The player learned the script, and the script did not learn the player.
Resource advantages substitute for reasoning
Higher difficulty settings usually grant the computer faster income, cheaper production or full map visibility rather than better decision-making.
This raises the challenge reliably and cheaply, and it changes the nature of the game, since the player is overcoming a handicap rather than being outplayed.
Players notice, which is why advantages granted this way are often disliked even by people who wanted a harder game.
Long-term planning is the hardest part
Strategy rewards plans spanning many minutes whose value depends on decisions made much later. Evaluating an intermediate position correctly is genuinely difficult.
An opponent that assesses only immediate strength will trade poorly and abandon positions whose value has not yet materialised.
Human players do this through experience and pattern recognition built over many matches, which is not easily encoded as rules.
Learning systems have different weaknesses
Approaches that train through repeated play can reach a high standard, and they require enormous amounts of computation and matches to do so.
They also produce opponents that are strong rather than interesting, since an opponent optimised to win does not necessarily create an enjoyable match.
This is why commercial games continue to ship rule-based opponents with adjustable advantages. The design goal is a good game, not the strongest possible player.