An intelligent agent has multiple goals, limited resources, and a dynamic real-time task environment. Because it has many predictable and unpredictable opportunities to act, the agent must "control" its actions-it must decide which of many possible actions to execute at each opportunity over a period of time. In a given situation, the agent’s control decisions determine whether it achieves its goals, what resources it consumes, and what side effects it produces. More generally, the agent’s approach to control determines the range of situations it can handle, its flexibility in responding to unanticipated conditions, and its coherence and comprehensibility to observers. How should an agent control its actions in order to achieve a high global utility?