Abstract
To improve competitiveness in complex and dynamic markets, many businesses seek agile partnerships. As a result, the paradigm of the Virtual Enterprise (VE) – a temporary network of cooperating enterprises in which there is a distribution of control, or decentralisation – has attracted much attention in recent years [1]. Successful collaboration demands a tight, but flexible integration of individual business processes across the VE, and a key element to achieving this is the use of workflow management systems to control these processes. Traditionally, workflow systems are based on two alternative approaches: model-driven workflow (e.g. [2]); and situated dynamic planning and negotiation between intelligent software agents (e.g. [3]). Each has strengths and drawbacks and some researchers, e.g. [4], have tried to get the “best of both worlds” by combining the two paradigms. However, these efforts largely take ad hoc decisions regarding the places and roles of agents in workflow management and construction. The approach reported here differs from these in the use of a formal model of the cognitive aspects of our domain constructed a priori. This allows us to determine the place of agents in a systematic fashion.