Abstract
A workflow involves the coordinated execution of multiple operations and can be used to capture business processes. Typical workflow management systems are centralised and rigid; they cannot cope with the unexpected flexibly. Multi-agent systems offer the possibility of enacting workflows in a distributed manner, by agents which are intelligent and autonomous. This should bring flexibility and robustness to the process. When unexpected exceptions occur during the enactment of a workflow we would like agents to be able to cope with them intelligently. Agents should be able to autonomously find some alternative sequence of steps which can achieve the tasks of the original workflow as well as possible. This requires that agents have some understanding of the operations of the workflow and possible alternatives. To facilitate this we propose to represent knowledge about agents' capabilities and relationships in an ontology, and to endow agents with the ability to reason about this semantic knowledge. Alternative ways of achieving workflow tasks may well require an adjustment of the original agent organisation. To this end we propose a flexible agent organisation where agents' roles, powers and normative relationships can be changed during workflow enactment if necessary. We use an example to illustrate how this combination allows certain workflow exceptions to be handled.