Any attempt to include learning competences in agent technologies, involves the problem of integrating such competences in the rationality of the agents. This work discusses the design of learning intentional agents in a multiagent system (MAS). Learning intentionally means that these agents learn to adopt their intentions under practical reasoning considerations, specified in terms of Beliefs, Desires, and Intentions (BDI). Specifically, they are intended to update the formula expressing when a plan is executable (context of plans), if some failure is detected. At the MAS level, a form of distributed social learning, based on collaborative goal adoption, is built around the concept of social awareness. Examples illustrate the role of induction of logical decision trees, an inductive logic programming method, in the design of these BDI learning agents