I really liked the way they grouped ships into task forces, with pickets, escort (?) and capital ships.
This concept really ultimately just combined ships into "bigger ships", functionally adding nothing. What happened when you built such a thing? Well, you had a single blob unit composed of smaller, no longer indidividually interactable, units. It was basically a big ship made out of smaller ships, no different than just having a single ship, except bigger.
Not really, if they can be targeted separately, and rearranged in several formations that would handle flank and rear attacks differently. The fact that it was badly implemented doesn't make it a bad idea. Actually, MOO2 made all smaller ships ineffective because of improved power/command point ratio, so there was no good reason to keep using smaller ships if you invested in larger hulls.
But I think handing micromanagement to the AI was misguided. It is much more efficient to abstract the things you don't want the player to five into.
You mean "dumb down the game"? Because that's the usual outcome of trying to abstract things, it makes the game dumber.
You still have to draw the line somewhere: Not abstracting things at all to avoid dumbing down would mean having stats for each individual recruits in your empire, and simulating each bullet individually. Abstraction is about removing factors that play next to no role at the scale the game is played on.
Otherwise, you end up with an unplayable mess like
The campaign of North Africa where there is no proof that a single group in the world ever played a full campaign, or Master of Orion 3...
The other issue with simulating everything at the lowest possible level of detail means that you have a much higher chance of doing your simulation wrong actually, because at its core, a wargame is a simulation in which you try to match the decisions the player make with the strategic outcome the designer thinks the tactical parameters should have at this scale.
If you have a bazillion factors involved, then it makes it much harder to refine the model because you cannot map decisions and outcome anymore.
Of course, overly absrtracting a game can also lead to bad results:
I really like Quartermaster General, because it lets you play WW2 in a single evening. Same for Twilight Struggle with the cold war, but a lot of the card based choices make no sense when trying to match them to real-world decisions.
