Despite the numerous findings on the sophisticated inferences that human infants draw from observed third-party helping interactions, currently there is no theoretical account of how they come to understand such events in the first place. After reviewing the available evidence in infants, we describe an account of how human adults understand helping actions. According to this mature concept, helping is a second-order goal-directed action aiming to increase the utility of another agent (the helpee) via reducing the cost, or increasing the reward, of the helpee’s own goal-directed action. We then identify the cognitive prerequisites for conceiving helping this way and ask whether these are available to infants in the interpretation of helping interactions. In contrast to the mature concept, we offer two simpler alternatives that may underlie the early understanding of helping actions: (i) helping as enabling, which requires second-order goal attribution but no utility calculus, and (ii) helping as joint action, which requires efficiency (i.e., utility) evaluation without demanding second-order goal attribution. We evaluate the evidence supporting these accounts, derive unique predictions from them, and describe what developmental pathway towards the mature concept they envisage. We conclude the article by outlining further open questions that the developmental literature on the interpretation of helping interactions has not yet addressed.