F.E.A.R. is a great case study in this respect, because the enemy AI is absurdly simple. Enemies only have 3 states, plus a plan. The neat trick was that inste

F.E.A.R. is a great case study in this respect, because the enemy AI is absurdly simple. Enemies only have 3 states, plus a plan. The neat trick was that instead of embedding the logic for changing state and setting parameters into the states themselves, they built a planning system which declares goals for the AI, and then allows the AI to procedurally generate the sequence of states it needs go through in order to be at that goal. To this day F.E.A.R.'s enemy AI is praised for being extremely deadly and realistic. Here's Jeff Orkin's GDC paper about it: http://alumni.media.mit.edu/%7Ejorkin/gdc2006_orkin_jeff_fear.pdf http://alumni.media.mit.edu/%7Ejorkin/gdc2006_orkin_jeff_fea...