There is a question that has quietly become the most consequential design problem of our era, and almost no one is framing it correctly: not what AI can do, but how we will tell it to.
For a decade, the interface was the product. UI was the mediating layer between human intention and machine execution — a surface we collectively obsessed over, pixel by pixel, tap by tap. Then the agents arrived, and suddenly the design community began whispering about screenless futures. If AI can act autonomously in the background, the argument goes, why do we need a screen at all?
But this is a category error. Eliminating the screen does not eliminate the interface — it relocates it. The question was never about glass and pixels. It was always about the grammar of human intent.
Consider what writing actually does. A PRD, a brief, a specification — these are not documentation artifacts. They are compression algorithms for thought. The act of writing forces the writer to resolve ambiguity before passing it downstream. The document is not the output; the clarification is. For decades, we outsourced our cognitive housekeeping to syntax.
AI dissolves that obligation. When the system is intelligent enough to tolerate ambiguity, to infer context, to ask a clarifying question rather than crash, the human no longer needs to pre-optimize their thinking. You can speak in the ragged, recursive, half-formed way that thought actually moves — and the machine meets you there. The PRD stage does not disappear; it gets internalized by the model.
This is why speech is not merely a more convenient input modality. It is a fundamentally different epistemological contract between the user and the system.
So the design question for the next five years is not screen or no screen. It is: under what conditions of life — walking, driving, building, caring for a child, sitting in a meeting — does a person need to externalize intent, and what is the lowest-friction, highest-fidelity channel for doing so in that moment? Sometimes that is voice. Sometimes it is gesture. Sometimes it is a glance at a wrist. Sometimes it is silence, and the agent infers from context alone.
The hardware form factor — phone, earpiece, robot, ambient room — is downstream of this question, not upstream of it. We have been debating the shape of the vessel before understanding what we are trying to pour.