An Experiment in the Design of Software Agents

Abstract

We describe a bottom-up approach to the design of software agents. We built and tested an agent system that addresses the real-world problem of handling the activities involved in scheduling a visitor to our laboratory. The system employs both taskspecific and user-centered agents, and communicates with users using both email and a graphical interface. This experiment has helped us to identify crucial requirements in the successful deployment of software agents, including issues of reliability, security, and ease of use. The architecture we developed to meet these requirements is flexible and extensible, and is guiding our current research on principles of agent design. 1 Introduction There is much recent interest in the creation of software agents (e.g., Etzioni et al [2], Maes [4], Dent et al [1], Shoham [7]). A range of different approach and projects use the term "agents", ranging from simple shell programming program systems to agents incorporating sophisticated real-time plannin...