J. Expt.Theor.Artif.Intell. 12(2000)191±212 Engineering characteristics of autonomous agent architectures PAUL SCERRI‹ and NANCY REEDŒ Real-time Systems Laboratory , Department of Computer and Information Science, LinkoX ping University, SE-581 83 LinkoX ping, Sweden ‹email: pausc!ida.liu.se Œemail: nanre!ida.liu.se Abstract . As the science of building agents and agent based applications improves, agents are gradually making the transition from research laboratory prototypes to industrial applications. In an industrial setting pragmatic engineering issues related to the development and maintenance of agents come to the fore. In this paper a list of important engineering characteristics of agent architectures is presented. The list has the dual aims of providing evaluation criteria for agent users and design issues for agent architecture designers. Evaluation of an architecture with respect to the presented characteristics will allow industrial project managers to better assess the trade oŒs between architectures. Designing architectures that better ful®ll the characteristics should lead to more industrially relevant architectures and successful agent applications. This list is designed to start discussion and raise awareness of industrial issues. It is hoped that over time good metrics and an evaluation methodology evolve which provide a robust framework for evaluating agent architectures with respect to industrial concerns. Keywords: agent architectures, designing architectures, chicken factory. 1. Introduction Intelligent autonomous agents are increasingly used in applications in many diverse areas. As autonomous pieces of software acting towards internal goal(s) within an environment, agents oŒer great promise as a means of increasing the utility of a complex system while at the same time reducing its complexity. Agents often encapsulate chunks of arti®cial intelligence within some larger application, allowing the development of intelligent, ¯exible applications from existing applications, as well as the development of new applications. As the science of building agents and agent based applications improves, agents are gradually making the transition from research laboratory prototypes to industrial applications. Successful application areas include control of complex systems, e.g. ARCHON (Jennings 1995, Cockburn and Jennings 1996) through to more abstract tasks such as scheduling or routing, e.g. (Chaib-draa 1997). Furthermore the applications to which agents are being applied are larger and Journal of Experimental & Theoretical Arti®cial Intelligence ISSN 0952-831X print }ISSN 1362-3079 online # 2000 Taylor & Francis Ltd http:}}www.tandf.co.uk}journals