In response to global competition, small and medium companies have begun to form alliances and networks between them. This allows each company to increase their ability to react and adapt to changes in their business environment, focus on their core competencies, increase the availability of resources and gain economies of scale. Thus, the model of fractal-based company based-project enables virtual and temporal integration between different companies in order to achieve specific business objectives. The successful implementation of the model lies in the establishment of clientserver relationships between project managers. This paper analyzes the integration of learning management agents to select partners with whom to establish the most beneficial client-server relationships. He advocates the application of reinforcement learning focused on the two-sided bandit problem.