The work presents a design and architectural perspective for the development of social robots with autonomous and context-sensitive decision-making capabilities, with particular reference to two application areas: artistic performance (MUSIC4D) and personalized support for healthy aging (MHARA). In both projects, the robot’s behavior is controlled by a structured knowledge representation that integrates affective perception, symbolic thinking and the generation of explainable actions. The aim is to promote forms of human-robot interaction in which the robot’s actions are understandable, motivated and socially relevant. A modular cognitive architecture is illustrated that combines consolidated technologies (ROS2, OWL, Neo4j) with advanced components for language management (LLM + LangChain) and emotional feedback processing. The two case studies show how the integration of symbolic models, inference tools and immersive environments can lead to dynamic and adaptive robot behavior. The paper is a position paper that aims to pave the way for a reflection on how knowledge design affects the quality of interaction and emphasizes the need for transparent, flexible and context-oriented architectures for the future of social robotics.