Some of the more complex AI testbeds are not that different from computer games. It seems that both sides, AI and game design, could profit from each other’s technology. We go a first step in this direction by presenting a very general agent control architecture (HAC: Hierarchical Agent Control), a toolkit for designing an action hierarchy. It supports action abstraction, a multi-level computational architecture, sensor integration, and planning. It is particularly well suited to controlling large numbers of agents in dynamic environments. We conclude the paper with an example of how HAC was applied to the game we use as our development domain: Capture the Flag.