In this paper we present a preliminary investigation of rational agents who, aware of their own limited mental resources, use learning to augment their reasoning. In our approach an agent creates and deductively reasons about possible plans of actions ,but — aware of the fact that finding complete plans is in many cases intractable — it executes partial plans which look promising. By doing so, it can acquire new knowledge from results of performed actions, which allows it to plan further into the future in a more effective way. We describe a possible application of Inductive Logic Programming to learn which of such partial plans are most likely to lead to reaching the goal. We also discuss how one can use ILP framework for generalising partial plans, thus allowing an agent to discover, after a number of episodes, a complete plan — or at least a good approximation of it.