This paper aims at analysing the effects of learning on the individual behaviour in an experiment that requires cooperation and coordination within teams. Using artificial agents, different social contests are created, as training environments. The results confirm previous findings (on the tendency to repeat the strategies that proved to be successful and to apply them also in other situations) and propose new hypothesis. Learning is not only based on a mechanical repetition of past choices, but also on reflection, imitation and on the attitude to build a model of the world. Besides the paper empirically tests the role of satisfaction in the routinization (this is one of the first empirical attempt of this kind). The training in an unfair and difficult environment leads some individuals to reinforce also strategies that have not been successful, but that involved emotions. Another contest seems not to determine detectable effects.