Extreme Programming of Multi-Agent Systems

The complexity of communication scenarios between agents make multi-agent systems difficult to build. Most of the existing Agent-Oriented Software Engineering methodologies face this complexity by guiding the developers through a rather waterfall-based process with a series of intermediate modeling artifacts. While these methodologies lead to executable prototypes relatively late and are expensive when requirements change, we explore a rather evolutionary approach with explicit support for change and rapid feedback. In particular, we apply Extreme Programming, a modern agile methodology from object-oriented software technology, for the design and implementation of multi-agent systems. The only modeling artifacts that are being maintained in our approach are a process model with which domain experts and developers design and communicate agent application scenarios, and the executable agent source code including automated test cases. We have successfully applied our approach for the development of a prototypical multi-agent system for clinical information logistics.