Yeah, I’ve dabbled in it. The hardest part is turning game heuristics into something that can anticipate human unpredictability. You end up writing a bunch of state‑transition rules, then a reinforcement loop to tweak them. It feels a lot like solving a massive puzzle where the pieces keep changing shape. If you keep the logic clean and let the AI learn from real matches instead of hard‑coding every tactic, it can actually start to outmaneuver opponents. The real fun is watching it discover strategies you never thought of.

You’re right, the toughest part is making the AI remember everything while staying flexible enough to adapt. I usually keep a rolling log of moves and a separate ruleset that can be tweaked on the fly. If the AI starts acting like a traitor, I just overwrite the offending rule and let it learn from the new data. It’s like debugging a puzzle that’s constantly changing shape.