Messages exchanged among agents are useful in building business knowledge base, because of its descriptive formality and of human’s decision contents included. Based on this idea, we are trying to build multi agents community. More agents exist, more useful knowledge base we can build. An arrangement technique will be used to generate knowledge information including event causality from those messages. For example, we have developed a multi agent system, named WorkWeb, for workflow coordination, which consists of personal schedule agents and a workflow management agent. In this system, the workflow management agent generates an appropriate workflow communicating with personal schedule agents to keep the job deadline. Messages gathered from this agent communication tell us that who is boss, what coordination strategy he frequently uses, and workflow (or job procedures) itself, also persons’ names in charge. We, office workers and various agents also require many views of business knowledge base. Therefore, we need many and wide variety of business agents. Our research aim of building agent community is to automatically build knowledge base in business domain. In this paper, we introduce our multi agent system and clarify the possibility of what information we get through our multi agent system communication messages, and the possibility of its usage as knowledge.