A HAII (Human-to-AI Interface) is the design discipline concerned with how a human extracts value from an AI system: the interface, the harness, the prompt scaffolding, the output rendering, everything that sits between the user and the model.
Why it's distinct from HCI
Classical HCI assumes deterministic systems: same input, same output, stable behaviour, sub-100ms latency. AI systems break every one of those assumptions: outputs are generated and variable, behaviour shifts as models update, latency is visible, and the system can be wrong in plausible ways. That's a new substrate, and it needs its own design grammar.
Why the term exists
The term was coined by OpenCues' creators to make the discipline investible: the industry defaults to a chat box because there's no shared vocabulary for anything else. Naming the layer, and evaluating it with tools like the 3-axis framework (start-up frames, active window duration, end-lag), is the first step to designing beyond it. The full argument is the HAII essay.
Examples
A chat window is a HAII. So are Copilot's ghost text, a voice assistant, and OpenCues' inline cues and blanks: different answers to the same question of how a person and a model should meet.
Related: What is inline AI? · What is OpenCues? · GitHub repo