Designing agent systems: State of the practice

Current approaches to the design and implementation of agents and multi-agent systems exhibit large differences both at conceptual and engineering levels. This paper argues that such differences result from different demands and characteristics of the domain. We substantiate this claim by a survey carried out online. The survey indicates that few agent systems are really built for industry. However, it also shows that the domain characteristics indicated have a high correlation to the type of agent systems that are used. Based on this evidence we classify agent systems using a few key components of agency. Using the survey, we give some systematic guidelines that support engineers on the assessment of the different approaches to choose an agent system type for a particular application domain. This forms the basis of a meta-methodology that starts from domain analysis and leads to a choice for the most appropriate agent methodology for that domain. Copyright © 2010 Inderscience Enterprises Ltd.