As I posted on another thread and as was alluded to be some posters here, the problem is not having AI behave smarter, but rather having AI behave realistically based on how they are defined in the world context.
For example, Raiders shouldn't be using a lot of tactics or strategy because they would not do such a thing. They do not believe in such approaches and only take the "might makes right" viewpoint.
The same thing applies to Super Mutants.
Ghouls (well, Ferals and their kin) do not have the brains to do such things so having them be "subtle" even in a context such as the Dunwich Buliding in FO3 would not match their given character concepts. Developers would have to make a different enemy, or at least a different type of Ghoul, in order to that to work (perhaps, in that context, have normal Ghouls be collaborating with Ferals in some way that is believable, although the latter caveat makes this difficult to swallow given the respective character concepts).
The above characters would not and should not behave in a coordinated way, either, at least in most circumstances, because they do not have communications devices or abilities in order to do so. Brotherhood of Steel or Enclave did, of course, so they should be capable of coordinated efforts. Raiders and Super Mutants might use some very crude coordination by shouting to each other, but again they really would not think of such things due to their straightforward viewpoint. Seeing AIs behave in ways that violate their given concept elements only makes combat a total pain because it violates role playing and immersion in the world.
Animals should pretty much never behave with human intelligence. Seeing animal AIs behave as a human would, including tactical use of cover and flanking strategies, is absolutely horrible for combat, although the latter could happen with packs of certain animals (if actual packs can be put on screen, of course). See the above observation about AIs behaving in ways that violate their concepts and how this is immersion-breaking and undermines role playing.
So sure, improvements are needed, but such improvements need to be very specific and not be something that is applied across the board to all AI concepts because each AI concept has its own defined behavior within the world.