Oh, yeah. I watch it think frequently. But to take that and evaluate it against another AI would take simulating the "battle engine".
OTOH, that the game does produce that kind of output suggests that hooks to replace the AI might be in place. It looks like the game evaluates moves (spell choices, etc) for each unit on the battlefield, and then passes them off to other routines to carry out. If the evalution and "do this move" calls can be brought out to an API, it shouldn't be too hard to attach Dom3 to another AI. (He says, knowing nothing of Dom3's internals.
)