The most basic improvement with this enhancement, adding to that, my leftover knowledge in programming, i could foresee that this only improves the bot's coordination by allowing them to be aware of the situation.
"Aggression"
This is the most basic thing, and we see this value tweaked based on pre-set conditions. To be honest, we really see a "thinking" bot as a whole. Individually, they are quite dumb magic shooter. But as a whole, they really keep their act together.
"Responsiveness"
Some of the ESAI codes involves the tweaks in bots respond time and its urgency in responding to the situation.
Try to be the commander sometime. Before this, i notice that the bots responded slowly to commands, some are even unresponsive. But responsiveness, based on my estimation, are 80% caused by navmeshes rather than AI codes. So even though i see improvement on my tested maps, but this doesn't instantly credits to the ESAI.
"Situation Awareness"
This is the main part of the enhancement. By making the bots understand that they have less flags than the opponent, they will know what to do. In some strategy sets, they will either try to counter attack (maintain flag balance, reduced aggressiveness), or just blindly push forward (attack to the last flag).
I wonder though, does it possibly allow the bot commander to defend a weak zone before enemy bots get within its cap area?
It really depends on where the bot spawned, as long as they have no order issued or instructed to defend the position.
At least the bots now man static guns, MG position, sit and wait along the defense lines. They really set up some "basic n00bish" defense. It is an improvement, but you really can't get the dynamics of human gameplay style. They don't respond to the flow of the battle, they simply receive orders and carry it to their death.
and, can bots be made to make a tactical retreat rather than bleeding away in trying to recap a flag? Then maybe countering with all their might?
Now this is can be done in some kind of a "save tickets" situation or any other conditioning you can think of. Because without a logic (if-then conditioning) to determine the situation, we cannot make the bot do the corresponding action. They just don't do unorthodox warfare tactics like humans do (ambushing, traps, coordinated retreats), which is all done using surprise element and the timing sometimes defy logic (which is why it is a "surprise"). Their decision making is determined by sets of logic, which made into several strategy sets made by Void.
One of the most basic example of the weakness is that, we at this point, still cannot make the bots to choose on which flag to attack other than "enemy" or "neutral". Which is then, depends on how we "neighbour" (link) the flags when we navmesh the map. They can't really choose to concentrate attack on certain points in a map. But then, they will assign each attack order per squad (divide manpower) if we choose to link all of them or it could be the otherwise if we artificially create a chokepoint by neighboring all of the flags into single flag as a chokepoint.
For example, in Totalize, we can't make them choose to attack Flak Batteries first than factory. What will they do is that, they will divide attack instructions per squads if we link the attack points from the main base to both.
Now that would make me do the jig
It is the ultimate goal of every AI developer. No that it is impossible or to sound negative, but I think you are really expecting too much from a 2006 AI engine.
My priority is, especially in FH2, to at least make the bots respond to our command. Just like the vanilla BF2 bots do.