This is similar to a concept I call "theaters", which I find useful when playing any 3X, 4X, Empire-like, boardgame, or other game with buildables and exploration (including some RPGs). Any map-based game with lots of things going on naturally demands a narrowing of scope (or you'll get overwhelmed by the complexity of interactions). Evidently, US and other militaries have encountered the same problem and found the same solution, for similar reasons.
Hence I advocate for a UI with "player-defined slices" of stuff, where a slice is both a map region and/or a subset of your units. A region is whatever you want to handle with a bit of tunnel vision, seeing everything germane to that region, and not-seeing anything else. Theaters evolve over time, because you eventually make progress (or give up), which changes your interest.
For the boundaries between adjacent theaters, maybe you simplify or abstract them away as (haha, here we go) input/output interfaces, or flows of stuff, or dependencies in a graph. Yes, it becomes a bit like OO-design with UML use cases I think that's not a coincidence. Humans simplify complexity, and it seems that we often do it the same way, across many domains and endeavors. Maybe hoomans are soft-wired that way.
Separately, the "theater" concept also makes sense for an AI architecture. A monolithic AI could do without it, and just try to solve the entire connected graph-of-everything every turn. A competing philosophy, which should appeal to *nix / agent / team programmers everywhere, is to model an AI as a generator-of-theaters, which spawns 1 "general" sub-AI for each theater. (I leave for future research the fascinating question of how the AI recognizes, pattern-matches, or deduces when to spawn a new theater, or to delete or merge an old one.) Thereafter, the sub-AIs compete for limited resources, and essentially float proposals upward, where each proposal demands some stuff, and promises some gain or effect. The "master AI" would then be like a CEO or king (or Joint Chiefs), and make grand strategy decisions for these-proposals and not those-proposals, presumably to maximize total (projected) return under the current constraints. This might not actually be the strongest AI in playing strength, but I conjecture that it would be the easier AI to develop. I think a human dev would go bonkers trying to implement the monolith as a monolith.
A theater-AI could also scale up nicely to handle the "panic response" in Empire-likes. The exemplar scenario is where a loaded troop transport or other threat suddenly appears off your coast. This demands a regional response, probably by diverting local assets and changing nearby cities to emergency-build armies or other defense. I think this could be elegantly handled by panic-spawning a new theater and general, and giving it extremely high priority, so that it can boldly make sweeping local changes, until the threat is handled. (N.B. this same architecture might generalize to any other kind of civilian or corporate panic-response, e.g. firefighting, or a Storm Surge/Flood/Tsunami expansion module for a SimCity-like, or net security vs. a hacking attack, or a Flash Mob Blue Light Special in a shopping-race game, etc.)