In this thesis, a distributed problem solving system, called MIAT(Multiple Intelligent Agent Testbed), and intelligent agent of basic computational unit of this system are studied. The architecture of MIAT has 5-level hierarchical layer: physical, distributed operating system, framework, intelligent agent, and user interface layer. Intelligent agent includes the facilities of adaptability, extensibility, and time-out. A framework in the framework layer of MIAT handles messages among intelligent agents, and schedules the intelligent agents. This system is applied for logic circuit design. Simulated results are compared with other works. By increasing information of other intelligent agents and problem environment, the number of communicated messages is less than that of communicated messages in distributed-processing system. Comparing with Yang & Huhns[21], furthermore, the number of controlled tasks is also reduced because in our model the planner updates the meta-knowledge when subtasks are dedued.