While the acquisition of knowledge about the sensory, emotional, and motor states of others has long been thought to derive from abstract inferential operations, recent theories based on neuroscientific and psychological models emphasize that motor simulation is essential to a variety of complex functions ranging from the monitoring of one's own performance to the understanding of others' actions, and matching self and others' states to optimize social interactions. In this review, we discuss evidence that simulating the actions of others has both bright and dark sides. On the one hand, we show that simulation can aid the anticipation and prediction of errors in the actions of others (e.g., in the case of competitive sports), as well as the establishment of social bonds (e.g., in the case of mimicry). On the other hand, based on findings from our and other research groups, we describe specific circumstances in which simulating the actions of others is detrimental to performance (e.g., when we automatically follow the gaze of a person who is actually trying to deceive us). Finally, we show how the presence of a shared goal between agents (such as in joint actions) maximizes the cost benefit of motor simulation, suggesting that the top-down modulation of this process is vital for adapting in a social environment.