The modality rule is a design principle for AI interfaces: any cue between a user and an AI system should use a different modality from the user's primary input channel. If the user is typing, the system's cue can be a visual mark or a note in the secondary display, never a popup that takes focus. If the user is speaking, the cue can be visual or haptic, never audio that competes with their voice.
Where the rule comes from
The Human Interaction essay derives it from how humans actually communicate. Two people speaking at the same time produce chaos, but one person speaking while the other nods produces steered communication, because nodding does not occupy the audio channel. The listener communicates continuously (nods, frowns, glances, backchannel sounds) without ever blocking the speaker. Every one of those signals rides a different modality from the words.
The same rule in software you already use
The essay's sharpest illustration is the WhatsApp reaction: a 👍 attached to a message steers without demanding attention, while the same emoji sent as a message would shift the chat window and interrupt the recipient mid-composition. Identical content, different channel, opposite effect. The pattern repeats across software: spell-check squiggles, IDE inlay hints, presence cursors in shared documents, typing indicators. Each lives in a separate visual layer, outside the user's primary input channel, and is fully ignorable.
What it rules in and out
| User's channel | Allowed cue | Violation |
|---|---|---|
| Typing text | Colour change, dim, underline, secondary-display note | Popup or modal that takes focus |
| Speaking | Visual indicator, haptic tap | Audio that talks over the user |
| Reading a response | A separate visual layer (reaction, margin mark) | Content injected into the text being read |
How OpenCues applies it
OpenCues' whole rendering strategy is the modality rule made concrete. Your words are the primary channel; the system's signal is a colour-channel dim layered on top of them, which never reflows text, never takes focus, and never borrows bold or italic (those already carry meaning in your writing). Tips render in the secondary display, outside the text. Even voice mode respects the rule: speech triggers only on navigation, when you have deliberately switched channels yourself. The essay states the payoff as the design test: the modality rule is what separates a cue that helps from a cue that interrupts.
Related: Prompt steering · What is an inline cue? · What is a HAII? · GitHub repo