The introduction mentions "industrial productivity" five times. I found it odd that it was not mentioned at all throughout the rest of the article.

      Sycara lists the types of knowledge used in design. Since Klein seems to be the first to suggest the use of knowledge about conflict resolution in his 1991 "Supporting Conflict Resolution in Cooperative Design Systems" paper, it is not surprising Sycara does not list this. I wonder if the system proposed would have been very different had Klein's idea been general knowledge in the field.

      "Design is the act of devising an artifact which satisfies a useful need, in other words, performs some function." - Rather than go on about the trouble with giving an absolute definition for design I'll simply state that many things have been designed that serve no useful function.

      A good discussion of how decomposition of a design problem effects efficiency was started in the 1.1 Design section. It gives a good basis for the extensive use of cased based reasoning throughout the proposed model.

      How cases used in this model are ordered hierarchically around concepts in the problem domain was not very well explained. It seems as though the hierarchy is simply function based where designs (full/partial) that provide some functionality can be found under the function description in the hierarchy.

      I found the constant use of in-paragraph lists: (1) annoying while trying to read, (2) confusing while trying to keep train of thought, (3) inconsistent, switching between numeric and letter formats to denote list items, and (4) amusing when Sycara gets confused herself and switches from numeric to letter format mid-list (see the paragraph before 4. A Negotiation Example). Inconsistency is one of the few consistent features of this paper.

      Inter-agent conflicts, requiring an agent to decide between tradeoffs, is mentioned but how this is handled by the model is not. It seems as though the same mechanisms used to negotiate between agents could be used internally.

      It is interesting how an agent is not ultimately limited by its local knowledge. An agent can learn about other agents in this model thereby expanding its knowledge to a degree based on the number of other agents it interacts with. This allows the advantages of distributing knowledge and enabling concurrent design that mimics human design teams while not imposing limits inherent in segregation of knowledge.

      At the end of the Agent Goals and Expertise section, Sycara states that goal traversal leads to innovative designs. I saw no further justification of this model being able to create "new" designs.