Using Agent Control and Communication in a Distributed Workflow Information System M. Brian Blake   View Paper (PDF)   Return to Agents Abstract Agent communication has developed widely over the past decade for various types of multiple agent environments. Originally, most of this research surrounded simulation systems and inference systems. Subsequently, agents are expected to adapt to, dynamically create, and understand evolving conversation policies. This concept of agent communication is not completely necessary in some domains, especially in domains where the policy of interaction is essentially static. One such domain is that of distributed workflow management with implications into Electronic Commerce. In this domain, agents are "middle-agents" that represent the distributed components that implement each individual workflow step. By representing the component-based services of each step, multiple distributed agents can essentially manage a workflow or supply chain that spans several on-line businesses (B2B). The WARP (Workflow-Automation through Agent-Based Reflective Processes) architecture is a multi-agent architecture developed to support distributed workflow management environments where distributed components are used to implement each of the workflow steps. This paper describes a software engineering process for integrating new component-based services into a static workflow-based ontology. Furthermore, the interaction protocol and supporting implementation based on the Knowledge Query and Manipulation Language (KQML) are presented. This agent communication architecture is implemented with the latest in Sun MicroSystems' Jini technology. DiSC'03 © 2003 Association for Computing Machinery