Interface primitives for agents
Interaction design matured around a human reading a screen. When the counterparty is a system that acts, the primitives — buttons, forms, confirmation — need rethinking.
Screens were the bottleneck
The vocabulary of interface design — the button, the form, the dialog, the infinite feed — was shaped by a hard constraint: a human, reading a screen, at a few hundred words per minute, with a narrow spotlight of attention. Everything was designed to route work through that bottleneck.
Systems that can parse, act, and respond dissolve the bottleneck. But the interface vocabulary has not caught up. We are still designing telegraph forms for a world that has moved past the telegraph.
Primitives, not products
This direction is about primitives — the smallest reusable units of human–system interaction. What is the equivalent of a button when the counterparty has intent? A delegation, a budget, a boundary? What does undo mean for an action taken on your behalf, minutes or days before you observe it?
We prototype small, concrete answers: an interface for granting a bounded capability rather than a password; a representation of what a system did while you were away, designed to be audited in seconds; a way to express a budget of trust — what the system may do without asking.
Design constraints
The constraints we hold ourselves to: interfaces must be legible without documentation; they must degrade gracefully when the underlying system is uncertain; and they must never use urgency as a design element. Confidence in a system comes from understanding it, and understanding comes from restraint in how it is presented.