Artificial intelligence is transforming assistive technologies, yet many systems remain reactive, fragmented, and insufficiently personalized for individuals with disabilities and neurodivergent conditions. This chapter examines the architectural, design, and governance foundations of trustworthy agentic assistive systems. It synthesizes limitations of conventional approaches, including restricted adaptability, low interoperability, and weak support for user autonomy, and introduces a taxonomy of agentic architectures to guide scalable deployment. The chapter proposes a multi-agent framework integrating multimodal interaction, hybrid reasoning, explainable AI, and health-aware data to enable context-sensitive support. It further addresses privacy, ethical governance, and responsible institutional adoption while outlining future research directions. By bridging architectural innovation with human-centered design, this work advances a forward-looking vision for inclusive intelligent environments.