@LorikEolmin
Well, the ATM frags based on the air picture at the moment as well as where are the main enemy airbases. It knows, as we know, where the enemy has its main squadrons. But it does NOT cheat by looking what the other side fragged or the enemy ATOs. The engine also does not remember playing this campaign before either.
There is one thing you guys need to remember: The engine must be able to deal any theater/campaign: for red and blue side. You guys have been flying these missions and campaigns for decades, so you know meanwhile very well what to expect. I bet many here know very the type of MIG in each enemy base for the KTO campaigns by heart. And where the CAPs are, right? Also everyone here knows where the SA-5 and SA-10s are exactly, right?
So, a human planning knows that all beforehand. The engine deals with the present picture (a bit in the future) and has (almost) nothing hardcoded. If Joe would go red side and play a different strategy for red (all manually controlled etc), the engine for blue side would still be able to deal with most of it.
On other side of the coin, I have experimenting with a deep-Q network trained to command the ATM/GTM. It is game changer so far in terms of fragging which could make the AI commander to behave much more human-like as you guys wish. But I digress.