This paper proposes a model for electronic institutions, promoting policy-driven autonomous service cooperation between different domains. The authors present an impressive and exhaustive definition of high-level infrastructure to model such cooperation using electronic business Extensible Markup Language (ebXML). The paper is well written and well structured, but from my point of view it has several drawbacks.

First, the authors’ approach is too abstract. They start with three main goals: “institution-governed cooperation, policy-driven self-management, and community facilitation management,” with multiple definitions proposed to explain them. However, it is unclear how these could effectively help agent systems. They should have adopted an explanation based on cases of use (such as the experiments mentioned on page 235, but not detailed at all) to show the benefits of adopting the proposed model. The authors, and interested readers, should look at papers from the Artificial Intelligence Research Institute (IIIA) about electronic institutions to see how this kind of model may be instantiated and used.

Second, the approach seems complex to understand, and complex to implement. Although the problem to be solved is challenging and not easy, the level of complexity implied by the application of this model would outweigh the possible benefits of using it.

Third, the approach seems to promote building agents ad hoc, instead of considering the applicability/integration of the proposal into any of the existing agent methodologies. The Foundation for Intelligent Physical Agents (FIPA) maintains standards to avoid these kinds of solutions. In section 6, the authors state that FIPA’s agent communication language (ACL) can be replaced by proposals like this one. I disagree with this point. Electronic institutions and the mechanisms involved to promote the appropriate use of policies have to be designed on top of FIPA standards, allowing their implementation with FIPA-compliant agents, not instead of them. Furthermore, it is not clear from the paper what kind of agents result from the proposed model. The authors describe a mental model of the agents, but do not specify if such agents are then deliberative and able to implement Rao and Georgeff’s basic reasoning cycle or not. The authors should have clarified this point.

Fourth, the assumed cross-domain feature mentioned in the title is not mentioned anywhere in the paper. If this issue is not relevant, this word seems inappropriate in the title.

Fifth, the mechanism to support the self-adaptation and self-evolution of virtual organizations, mentioned in the last paragraphs of section 5.2, is interesting enough to be discussed in its own paper. The authors could have extended this part to show its real applicability with a use case.

Finally, some minor errors in the references should have been fixed: all the authors of a paper should have been listed, instead of using “et al.,” and references retrieved from the Web should have included the date the information was accessed.

For all these reasons, the paper is not as effective as it could have been. That being said, I think it would benefit people from the agents community.