This work focuses on the importance of exploring how people explain and assess the behavior of artificial agents in social contexts. It introduces a formalization of agent behavior through a cognitive architecture based on Social Practice Theory. Using narrative scenarios enacted by the humanoid robot Pepper, two distinct behaviors are compared: one that is socially evocative and another that aligns with social expectations, driven by the interaction of the components within the architecture. A preliminary analysis combining thematic coding of semi-structured interviews and a Likert-scale questionnaire reveals that when a robot’s behavior aligns with social expectations, it influences both the language used to explain the agent’s actions and the participants’ assessments of its social competence.