This reminds me of a programming assignment from college to write a chess bot.

My partner and I tweaked it's understanding of how to evaluate board positions to eek out better playing ability, but the major improvements came from implementing a more efficient specialized game tree that let us calculate another move ahead every decision.

Having one more "loop" made a huge difference against our opponents; being simply smarter or better at any of the "steps in the loop" didn't help nearly as much.