AI language models are transforming how people communicate, offering unprecedented speed and efficiency. For users of Augmentative and Alternative Communication (AAC), who rely on typed messages to speak, these technologies hold particular promise but also pose unique challenges.

In this talk, I explore how AI can support, reshape, and sometimes complicate communication for AAC users. What happens when systems begin to anticipate or generate what someone might say? How do we ensure that AI supports expression without overriding the user’s voice? Drawing on human-centered design, participatory research, and systems developed in our lab, I share examples of emerging AI-supported AAC tools and design explorations created with people with speech and motor disabilities. These projects surface key tensions between speed and control, assistance and authorship, automation and identity.

I conclude by outlining open design directions for building AI systems that work with AAC users rather than simply for them, prioritizing transparency, adaptability, and user steerability. Ultimately, the goal is not only more efficient communication, but technologies that expand people’s ability to express themselves on their own terms.