Fair point. However, I think that the core problem is that the lack of stacking is simply a design flaw - in the sense that it's too complex for a small team of programmers to write good tactical AI for the Civ setup (e.g. no fixed maps, and a wide range of possible unit combinations.) I can't see them abandoning the combat model in any Civ 5 expansion, so fixing the core complaint about the game would be very, very difficult to do. Variants of it could do quite well in future versions, e.g. with a tactical map (no stacking) combined with strategic level stacking and possibly an army concept.
I think this is a reasonable point, but at the same time is the combat system the key problem, or is the AI's willingness to use a combat system it's unsuited for the key problem? If the AI declared war less frequently, its inability to wage war would be a lot less obvious. True, aggressive player strategies would still be rewarded more than they should be, but little is quite as boring as being forced to continually defend against piecemeal attacks by an AI that can't fight, or to see an enemy declaration of war against you as a good thing because it promises to throw away a lot of resources.
Beyond that, I think there are certainly combat improvements that can be made which aren't so context-dependent as your suggestions. Cities, wherever located, are static targets - there are standard strategies for attacking them which the AI is at best intermittently capable of implementing. I can lose a city defended only by a ranged garrison unit to a Japanese attack that uses a lot of archers and a final push by a couple of Spearmen, but in the same game defend another against an unending horde of Mongol Warriors who never thought to bring bows to the party.
People have often observed that ranged units and siege often lead in AI strategies; Civ V's division of units into several key unit types (cavalry, spears, melee, ranged, siege) should make it easier to design algorithms dictating how these units should interact, what 'counters' what (how many times have I seen the AI waste cavalry against my pikemen?) etc. This, I presume, is how the more capable AI in Total War games makes decisions - and it will send spears specifically to head off a cavalry charge, and will usually defend its missile troops.
When defending a city, the AI should know to prioritise destroying the units that can actually take the city where possible, rather than just targeting whichever ranged unit dealt it damage last turn as seems to be the case now.
Terrain has fixed effects that the AI essentially seems to ignore - it will quite often attack across rivers when it has otherwise equal units, and will lose as a result.
A personal bugbear: Teach the AI to recognise how Citadels work, rather than having units that it calculates can't attack stand around like lummoxes taking damage. Related to this, it appears that the AI usually operates by calculating odds on a unit-by-unit basis - it will often stand around and not attack if it has three units surrounding yours, because none of those units can individually expect a victory. In the same situation, a human player would recognise that three attacks together would wipe out the enemy unit.
And one of the simplest of all: use unit upgrades. I don't think the AI ever does this, nor does it seem to recognise when a unit becomes obsolete - it will only start producing modern units when old ones become unavailable. It relies heavily on producing UUs, regardless of context or time period. In my current game it's the early 19th Century, and both I and the Arabs have riflemen. The Japanese have a high score to a giant empire, but they're still attacking with archers and Samurai. I wonder if the AI actively avoids researching technologies that will obsolete its favourite units? I'm sure I've played against Babylonians who don't seem to want to research Engineering before now.
