ai-native
AI-native design specifies interfaces whose screen, affordances, and content are produced at runtime by a model rather than authored ahead of time — agent UX, conversational and agentic UI, and generative non-deterministic UI. There is no settled canon for this pillar. Unlike the other design pillars (each anchored in established books), this one is principle-derived: drawn from practitioner essays and emerging sources, marked as such on every claim. Cite the principle and the canon gap.
When this applies
- Designing an agent UX / AX surface — a system that takes goals and acts, rather than rendering controls the user operates step by step.
- Designing conversational or agentic UI — chat, tool-calling assistants, multi-step autonomous flows.
- Designing generative / non-deterministic UI — where the model assembles the layout, copy, or component set at request time, so no two sessions are identical.
- Building a no-fixed-screen interface — no stable set of pages to map, no fixed affordances to learn.
- Calibrating trust and control for AI output — confidence display, undo, correction, escalation to a human or a deterministic path.
- Diagnosing which classical pillar strains when AI removes its fixed substrate, and what to substitute.
Not for: whether a conventional screen is operable — heuristic evaluation, UX laws, affordances on a fixed UI (use usability); the visual DNA, fonts, or color (use core design); a voice or conversational surface treated as a device class with its own layout constraints (use core surface).