Training teams increasingly takes place in synthetic environments. However, team training is often still modeled after live team training, including the disadvantages of live training, such as instructor-intense performance monitoring, and the fact that all appropriate other teammates have to be available. This paper explores the latter issue: how to overcome the bottlenecks of the availability and drawbacks of human teammates in training teams in synthetic environments, while keeping the advantages: the opportunity to learn in a collaborative and cooperative fashion. Simulated teammates are a promising alternative to human teammates, because they are always available, may be modeled after experienced training personnel, and may be more cost effective in the long run. The research challenge lies in keeping the advantages associated with human teammates: simulated teammates should display the same collaborative and cooperative behavior typically associated with human teammates. This paper will review the relevant available research data, and will explore how intelligent teammates should be defined and modeled so as to take advantage of both worlds: optimizing the possibility of cooperative learning, as well as optimizing individual and team learning experiences.