I don't work in the gaming industry, but it feels to me that engineering a bot like this is a dead art going forward, and that makes me sad. Looking forward (today?) I assume this will be solved by training a black box neural network. Iirc In the AlphaZero paper they train bots to play chess against each other, and thus pull the thing up to 3000+ ELO by its own hair. I guess the same works for FPS bots, and you can save different ELO strength bots, so different level players have enjoyable bot opponents?