The dominant metaphor in interface design is the container: a box that holds content. Menus, cards, modals, drawers — these are all containers. The content goes in; the user takes it out. This model made sense when content was authored by humans and delivered at human pace. It makes less sense when content is generated in real time, shaped by context, and potentially infinite in variation.

AI-native interfaces need a different metaphor. Not the container, but the environment: a space that responds to the intelligence within it rather than constraining it. An environment does not have fixed slots for content — it has properties, behaviours, and affordances that shape how content manifests without determining what it will be.

This is not a small distinction. Designing a container requires knowing, in advance, the shape of what will go inside it. Designing an environment requires understanding the properties of what will inhabit it — its density, its rhythm, its variability — without predetermining its form. The second kind of design is harder, and more interesting.

In practice, this means building interfaces that can hold sparse content and dense content without looking broken in either state. It means designing for streaming: for content that arrives incrementally, that expands and contracts, that may need to be corrected or extended after the fact. It means treating uncertainty as a first-class design condition rather than an edge case.