There won’t be a Skynet, at least not a widespread one.
What would WOULD like to do is implement a lot of smaller, simpler, but significantly less stupidly insane AIs that have good reason to be hostile to the player.
I mean, they will still probably be insane, you know, by our standards, but realistically so, not like with the SkyNet model of “This computer has spontaneously decided to be a dick and is also kind of dumb”.
We’ll also be having various types of controller AI as well, of varying complexity, and obviously many of them will have quite limited reach with the global networks all done (although most of them would have been fairly constrained anyway). And they won’t necessarily all be friendly with each other.
The Primary AI types will be Lab AIs, Corporate AIs, Security AIs, Military AIs, Service AIs and Instance AIs.
Of these, the Lab AIs are likely to be the least stable, since many will be experimental, the Service and Security AIs the simplest and least versatile, the Corporate AIs the most useless and the Military AIs the best armed.
Instance AIs will vary the most widely, are the type of AI the player can build right now (stuck in their turrets to make the magically aim and detect enemies). They tend to be little more than a set of behavioural rules though.
I’d say the Lab and Military AIs are the best options for awesome gameplay stuff, with the Corporate AIs being potentially amusing.