The AI is more situation based than map based although each scenario has map specific extra AI embedded in the scenario file. Most of the AI is in the game engine, some is embedded in each scenario, there is none embedded in a map. The AI is not really hardwired for exact units. It looks more at the class and weapon systems of its units rather than at the fact that this unit is a T80 and that unit is a T72.
The AI is both scripted and reactive. Barring interference from its human opponent, it follows a script or a battle plan that will accomplish its scenario mission. If the human interferes with the plan (i.e. gets in the way with units and or fire) the AI usually switches to reacting to legally acquired spottings of the human's units - sometimes it just ignores minor spottings and or treats them solely with arty. If there are several minutes of no human interference, the AI will usually switch back to a script. A script or battle plan may include a general movement path, precautionary assaults on intermediate terrain objectives (even if there has been no contact), precautionary pauses followed by recon, rolling arty prep fires, etc. Reaction to contact varies depending on wether the AI is attacking or defending but can include ignoring minor human resistance, frontal assault from the column, deployment to a frontal online assault,deploy and maneuver to a flank assualt, deploy and maneuver to a rear assault, pause in place and shell, split a force to do combinations of the preceeding, temporary withdrawal (not seen very often when AI is attacking), counterattack, online sweep, recon by fire, recon in force, fire and move to alternate position, hasty defense, etc.
Summation: the AI chooses its basic battle plan from its bank of multiple opening moves right after the OPFOR order of battle is loaded and before you have the chance to do a save game. As the OPFOR units make contact with US units, the AI task organizes its battalions and or companies and switches to reacting tactically to those contacts. As it successfully deals with contacts, it switches back to the primary battle plan or it may jump to a different battle plan.
The AI has the same time limits as a human player has. Here is how the AI does its thing with calling in air strikes (arty is handled similarly). During each combat phase the AI makes mental notes about points on the map where it legally spotted your units --- same as a good human player does. In the case of moving units, it also projects and records where the unit could be in a minute or so based on its speed and its last legally observed direction of movement. I call this maintaining a target list.
During the next orders phase the AI plans its orders based on its target list, again based on what it legally saw during the previous combat phases --- same as the human player does. The first time it spots valuable units it will probably target some or all of them with air and arty up to the limits of its legally available assets. AI OPFOR air gets no time advantage over a human player. During the four to eight combat turns that must pass before the AI's air arrives on the map, the AI continues maintaining and considering its target list of legally spotted units. The AI is allowed to shift ongoing air and arty missions the same distance per turn as the human player is allowed to. Thus as time passes and the AI gains and loses legal unit spottings and as more valuable targets become noticed, it often modifies its air and arty targeting with legal target shifts. Enemy helos are the highest possible priority target for AI air support. Anytime a US or OPFOR airstrike comes on the map it looks for helos within 1000 meters or so of its target point. If it finds a nearby helo, the airstrike will usually switch to a gun attack on that helo. Usually the air unit will prevail over the helo target. Usually an air strike that has diverted to helo attack will also remain available for reuse. By the way, human players also get these very same helo attack benefits with their air.
Don't try to figure out the rules and the computer logic of the AI and learn the grim tactical lessons it is teaching. Develop sound real world counter tactics rather than computer tricks. Treat it as prudently as you would treat a human opponent and you should have great success.
The observation that OPFOR units are not firing at you at a given point in the game does not necessarily mean that they are not quietly and legally watching you from a hidden position.
If there really has been no contact and if none of the OPFOR units are watching or have recently spotted your units then the AI does not know specifically that you are in the places being hit by its arty. What is happening is that the AI is shelling places that "would" be good places for you to be at that point in the game. If you watch the impacts, you will usually see that they wander around a lot from volley to volley until maybe a secondary explosion occurs. At which time the AI may then concentrate fire on that point for some minutes.
There is some AI code embedded in each scenario that is specifically designed for that scenario. Part of this special AI code is a "target list" of places on the map that would be very good places for the human player to put his troops. If a terrain feature looks good to you as a fighting or a hide position, it may also look good to the AI. For example, a clump of woods very close to a position defended by OPFOR might be a good place for the US player to launch his attack. That clump of woods may be on the AI's arty target list. If the AI does not have anything better to do with its arty,it will often choose to target these places --- same as a good human player would. So it is possible for you to do everything right in carefully occupying a position without being spotted and still find yourself later under air or arty attack. In such a case the computer is not cheating, it just made a good or lucky guess.
The computer opponent gets an initial, one time, advantage of one accuracy level for observed arty fire when it marks a new arty mission. However, it gets no extra advantage in killing effect of impacting arty. In other words, the computer opponent uses exactly the same combat results tables as would a human playing OPFOR.
The priority of the TacOps AI is not to beat you into the ground or to try to always win. The primary purpose of the TacOps AI is to provide you with a computer opponent that does its thing using fairly realistic tactics. If I cared about the AI always winning there are any number of "game tricks" that I could let it use, but I don't so I don't <grin>.
Coding AI is hard unless it is ok with the author for the AI to be stupid and or to cheat. These things are not ok with me so I find all AI to be extremly dificult.
However I have found that tolerable offensive AI is somewhat less hard to write than tolerable defensive AI. By tolerable I mean that the AI is less limited to having to do the same thing every time and thus is less predictable. A real world problem with being on the defensive is that it makes your troop dispositions more predictable.
Standard scenarios have a big part of the computer opponent imbedded in the scenario file, so that the AI can be modified to take maximum advantage of the missions, terrain, and situation for that scenario. Custom scenario templates have no constraints, no boundaries. Unless I let the user program the AI --- a bad idea since that would spoil the play of the game for the creator --- the computer opponent for such games would tend to be consistently weak. I would rather provide no AI than to provide bad AI.