The advent of inexpensive computers has spurred interest in distributed architectures, in which a cluster of low-cost cornpeters can achieve performance on the same scale as expensive super-computers. The goal of research in distributed systems is to take a highly decomposable problem solution, and put that solution into a framework that allows the system to take full advantage of the sohition’s decomposability. An architecture that allows additional processors to be taken advantage of with little or no manual reconfigurafion would be ideal. Artificial Intelligence (AI) algorithms for planning in nontrivial domains am typically resource-inteasive. We believe that a framework for developing planning algorithms that allows an arbitrary level of decomposition and provides the means for distributing the computation would be a valuable contribution to the Al community. We propose a general, flexible, and scalable hierarchical architecture for designing multi-agent distributed systems. We developed this architecture to facilitate the development of software for soccer-playing robots. These robots will compete in the international robot soccer competition RoboCup. All of our software, from the high-level team strategy planning down to the control loop for the motors on our robots fits into this architecture. We build on the notion of deliberative and reactive agents, forming a hierarchy of hybrid agents. The hierarchy covers the entire continuum of hybrid agents: agents near the root of the hierarchy are mostly deliberative, while agents near the leaves are almost purely reactive.