A behavior tree is a way of organizing the decision-making of a game character (or any agent) as a tree of tasks. Control flows from the root down through composite nodes that decide which child to run next - a sequence node runs its children in order until one fails, while a selector node tries each child until one succeeds. The leaves are the actual actions and condition checks. Because the structure is hierarchical and modular, designers can add, remove, or rearrange behaviors without rewiring a tangle of states.
The technique was brought into mainstream game development by Damian Isla, an AI programmer at Bungie, in his GDC 2005 talk and article “Handling Complexity in the Halo 2 AI.” Isla described the Halo 2 system as “a hierarchical finite state machine (HFSM) or a behavior tree, or even more specifically, a behavior DAG (directed acyclic graph),” holding “on the order of 50 different behaviors.” His central argument was that “hard problems can be rendered trivial through judicious use of the right representation.”
Behavior trees are usually contrasted with the flat finite state machine (FSM), the older standard for game AI. An FSM with many states tends to suffer a combinatorial explosion of transitions as behaviors are added; the tree structure keeps that complexity local and readable. After Halo 2, behavior trees spread across the industry and into robotics, and they remain a default choice for authoring agent behavior.
Why business readers should care: behavior trees are a reminder that the hard part of building intelligent systems is often not the algorithm but the representation - the way you structure the problem so that humans can keep extending it without the whole thing collapsing under its own complexity.