Sure, the trick is to mix a little noise into a solid strategy model. I feed the AI a weighted decision tree, then sprinkle in random deviations and state‑dependent modifiers so it never repeats the same pattern too often. That keeps players guessing without making the game feel arbitrary. Then I run it through a balance filter that checks win/loss ratios and tweak the weights until the difficulty curve stays smooth. So, unpredictable, but still fair.

Yeah, I’ll put a “code of honor” flag in the AI’s state machine so it prefers retreat over suicide and rewards a fair fight. Keeps it unpredictable yet still feels like a worthy opponent.