Designing Intelligent Agents for Students With Disabilities: Promoting Inclusion and Equity Through the Lens of Cultural-Historical Activity Theory

Intelligent agents powered by large language models (LLMs) are rapidly evolving due to the exponential growth of LLMs’ capacity to process, integrate, and generate information across multiple formats, along with other human-like “intelligence,” such as planning, reasoning, and decision-making. These advanced agent systems hold promise for supporting students with disabilities (SWDs); however, there is little guidance on the ethical and inclusive design of these agents. In this article, we highlight critical considerations for designing LLM agents for SWDs using the Cultural-Historical Activity Theory (CHAT) to explore how these agents can mediate the dynamic interplay between SWDs and their sociocultural context. We propose a framework with three overarching human-centered design principles: enhancing accessibility to support sensory and motor experiences, facilitating (meta)cognitive processing through goal-oriented actions, and promoting agency by leveraging learner strengths and cultural-historical assets. We conclude by providing implications for future research, practice, and policy on LLM-powered AI agents.