Four experiments investigated whether infants and adults infer that a novel entity that interacts in a contingent, communicative fashion with an experimenter is itself an intentional agent. The experiments contrasted the hypothesis that such an inference follows from amodal representations of the contingent interaction alone with the hypothesis that features of the experimenter's behavior might also influence intentional attribution. Twelve- to 13-month-old infants and adults observed a novel entity respond contingently to a confederate experimenter, the form of whose actions varied across conditions. For infants, intentionality attribution was assessed by the extent to which they subsequently followed the faceless entity's implied attentional focus. For adults, intentionality attribution was assessed from their use of psychological terms when later describing the entity's behavior. In both groups, construal of the entity as an intentional agent was limited to a subset of contingent interaction conditions. At both ages, the pattern of responses across conditions suggests that whether an observed contingent interaction can be seen as a social interaction influences the attribution of intentional agency. These results further indicate that the agent detection mechanism responding to third-party contingent interactions, as a context-sensitive process, is distinct from the mechanism responding to directly experienced contingent interactions, suggested by prior developmental work to be based solely on amodal representations of an entity's contingent reaction to behaviors of an infant.