Automating static cross-organizational processes is already difficult, but this paper tries to address the additional complexity of dynamic processes, where neither process flow nor (all) process actors are known at the time a process starts.

Sadly, however, the technical content and the actual realization of the approach by the authors in a working system remain fairly disconnected. The architectural recommendations are essentially limited to specifying the different roles of agents required to realize such a system (three for workflow enactment and ten for connecting with clinical decision support and computerized prescribed order systems). The authors do provide a lot of text and graphics to show fairly trivial portions of their architecture, but the more difficult (that is, technical) parts of the design (for example, interactions between the different agents) are kept to the level of acronyms, for example, agents shall use the agent communication language (ACL) advocated by the Foundation for Intelligent Physical Agents (http://www.fipa.org/).

Contrary to the title, the medical setting is not crucial and the work can be easily generalized to other industries. Furthermore “web services” technology is reduced to a rather unexplained set of standards that can ease the integration of heterogeneous distributed applications. The fact that the authors suggest using the now-obsolete BPEL4WS language for web services orchestrating is ill advised, and organizations should instead focus on business process model and notation (BPMN) 2.0 as the de facto standard in this area.

Additionally, even though the authors (also correctly) recognize that the 1994 reference architecture of the Workflow Management Coalition (WfMC, http://www.wfmc.org/) is fairly outdated in general and for this purpose in particular, they seem to continue to (wrongly) use the now-obsolete interface specifications thereof. Thus, they even fail to recognize that WfMC interface 3 (workflow engine calling external systems) is superseded today by the very web services the paper rightfully advocates.

Because the paper has been published in a journal covering medical systems, technical concepts are explained very elaborately (perhaps even repetitively) and remain on a highly superficial and nontechnical level. If you really intend to build such a system, you will not be able to do so by just reading this paper.

To summarize: unless you are really interested in agent technology and how many different roles of agents one needs to solve the dynamic workflow coordination use case (which in principle is interesting), you can ignore this paper.