A distinguishing characteristic of Subsumption Architectures is its lack of a world-model. Brooks sees this as an advantage. By not having representation, the architecture avoids the time necessary to read and write them, the time-cost of algorithms that utilize them, and the difficulty of maintaining the accuracy of world-models. These are all problems that are major issues in any system with a world model which also attempts to operate in complex and dynamic environments.
But this policy is a double-edged sword. True: such systems are fast. But they also depend entirely on the world to tell them what to do: they are purely reflexive. Behavior-based systems (Mataric, 1992) attempt to add some representation without losing the robustness and reactivity of pure subsumptive architectures.
Go to a discussion of this property for multiple architectures.