When Agents Clash

The notion of agency is central to the study of software process. Software development processes are enacted by agents, represent agents, and ultimately, must be capable of guiding human agents. This paper explores the use of the term “agent” in software engineering, and advocates a systematic study of this notion in the context of software process technology. In particular, the paper proposes the explicit use of agency concepts in order to support the management of conflicts that inevitably arise in software development processes and products. 1. Agents in Computing By far the largest body of work on agents in computing emerged from work on (Distributed) Artificial Intelligence, and more recently Multi-Agent Systems and CSCW. As a result, the most common use of the term agent is as an entity which exhibits some kind of intelligent behaviour or “attitude”; e.g., has beliefs, desires, etc. Moreover, agency appears to be characterised by properties of autonomy, reactivity (e.g., perception), proactiveness (e.g., taking the initiative in a goal-directed way), and social ability (e.g., ability to interact) [6]. In this context then, software agents normally refer to processes that users may delegate tasks to, and that may require yet other facets of intelligent behaviour such as planning, communication, co-operation and decision-making. The term “process” in the above is used in its most general meaning (as an ordered set of activities that achieve a task), but as this paper will argue, it may also be used in the context software process technology in particular. A variety of software agents have been described in the research literature [4]. Some commonly known ones are interface agents that provide assistance to users of particular applications, autonomous agents that work on behalf of users without requiring any particular interaction with those users, information agents that access and manipulate information in response to user or other agent requests, and intelligent agents that show “emergent” behaviour; e.g. creativity, common-sense, etc. 2. Agents in Software Engineering A working definition of the term “agent” in software engineering is as an actor with a particular role in the development process typically a development participant with responsibility for achieving a particular development task. However, while the software engineering community has examined the separation and relationship between the * (to appear in) Proceedings of 5th European Workshop on Software Process Technology (EWSPT96), Oct ober 96, Nancy, France, Springer Verlag.