Designing new agent-based applications architectures with the AGP methodology

Many fields of research in Computer Science use the concept of software agent either as a tool, or as a concept: artificial intelligence, artificial life, distributed systems and applications, applications related to the e-economy, concurrent engineering, etc. By analyzing two applications domains, multiagent systems and distributed applications, we shall show that the same methodology, called APG methodology (APplication first, aGent after) is mainly used. This one leads the designer to define at first the characteristics of the application, then those of the agents which will implement it. The major drawback of the APG approach is to limit the expression of the properties of the agents to a particular realization of the target application. We shall describe the reverse approach called AGP methodology (AGent first, aPplication after) which consists in adapting preexisting agents to the implementation of the target application. The choice of an agent model is made with regard to its profile. In short, a profile defines the set of properties which the agent possesses. AGP approach is based on a clear classification of the agent properties in two categories, qualities and attributes, and their distribution in two subsystems: those which are native with the reused agent and those which must be added to it or modified so that it can completely carry out the new application. The use of the AGP approach will be illustrated with a new class of agents: systemic agents.