This is an interesting idea but i suppose it depends on whether we are simulating our place as a GM or a coach. It seems we are actually skirting the border between the two, taking on some functions of each.
A better, smarter coach should react and implement changes more intelligently, while during a game i would suspect a GM should not interfere much more than giving the coach whatever leeway in taking risks to 'win this one'.
In practice, though, i would think that the AI is feeble enough that it shouldn't be given more rules to follow and decisions to make than it already has, and the limit of GP changes probably should be specified beforehand by the human AI, such as it is presently.
