Because the execution of many complex tasks is increasingly relying on human-agent teams it is critical that we understand the processes arising from such interaction and the specific conditions affecting them. Despite findings surrounding effective interaction and coordination for expert teams in general, little is known about what is important in expert human-agent teams. Paramount to the effective utilization of human-agent teams is the appropriate blend of research to investigate the boundary conditions within which training must be tailored and delivered. In this paper we describe a representative framework through which the research community can investigate human-agent teams. Our framework involves a blending of laboratory and field research methods with training research. We describe how coordination demand analysis in conjunction with lessons learned systems can be used to capture critical incidents and data from expert human-agent teams performing in context and how this information can form the foundation for effective human-agent team training.