Abstract: Attention is being paid to the development of large and complex multi\|agent systems. A theoretical framework of agent computing is put forward in this paper to offer method and concepts to specify the multi\|agent system’s requirement, and to direct system’s design. A new architecture （BICC） of agent is brought forward, in which agent is an intentional system composed of belief, intention, capability, and commitment. Based on the logical framework of multi\|agent computing, the semantics of each cognitive component in the architecture is defined and how agent computes based on the architecture is investigated. Finally, the live and safety properties of agent computing in a dynamic and non\|deterministic multi\|agent system are proved.