Self-adapting agent organizations have been promoted as a useful and suitable paradigm for open and dynamic agent systems. This paper argues for a flexible approach that provides a meta-reasoning component to agents using localized adaptation rather than a centralized mechanism. This novel approach is presented for meta-reasoning using social utility factors to help decide whether or not to adapt an agent organization.

Organizational structure is defined in terms of agent relationships. Changing the organizational structure entails changing, dissolving, or adding localized individual agent relationships. In this paper, the organizational structure is implicitly represented in the individual relationships, not explicitly represented at an organizational level. Each agent may have superiors, subordinates, and peer agent relationships. Subordinates or peer agents provide services that can be allocated. There is no first-class agent organizational entity as such.

To address the dynamic nature of organizations, the authors propose a weighted (decaying with time) factor to calculate the utility of the relationship. The authors present three algorithms to apply based on the type of organization. The fundamental algorithm, k-adapt, is extended for open organizations based on the WoLF principle: “Win or Learn Fast.” In other words, if you are working well with spare capacity, don’t change quickly; if not, restructure quickly. For dynamic organizations, the k-adapt algorithm is extended to include decay over time, so that more recent relationships have more importance.

The paper includes a modest but useful review of related work. It will be of interest to both those who want a sense of why and how to study self-organization from a multiagent systems perspective, and those who are interested in the fine details of a carefully addressed approach in one particular setting.