# Generative UI

DS230. Future / AI: Runtime Composition. Page: https://ui.codeatlas.com.br/atlas/generative-ui/. Index: https://ui.codeatlas.com.br/llms/atlas.txt

The UI is assembled or transformed for the task from context and intent.

- Status in 2026: Emerging. Appearing in practice, not yet settled.
- Trend 2027 to 2030: Potential (an editorial projection, not a forecast).
- The term: Known technical term or standard.

## Traits

- components on demand
- adaptation

Typical settings: Copilots, Search, Analytics.

## When to use it, and when not

Consider it when the task and motivation match: users with varying questions and flows. Typical settings: copilots, search, analytics.

Be careful when the risk below would hurt the main task: inconsistency, accessibility and trust.

Audience (a hypothesis about motivation and task, never measured, not a statement about any group): Users with varying questions and flows.

## UX risk and how to reduce it

- Risk: Inconsistency, accessibility and trust.
- Mitigation and accessibility: Validated components, limits and inspection.
- How to check: Test with realistic scenarios: task success, accuracy, reversal, auditability, understanding of what the system does on its own, and a human's ability to step in.

## Evidence

Emerging: exploratory work, with no causal validation for this style.

The sources support the principles that apply to this term. None shows that this style works better or is preferred.

- [E17] Lee (2025). Towards a Working Definition of Designing Generative User Interfaces. https://arxiv.org/abs/2505.15049 (Preliminary). Does not support: Effectiveness of generative UI. It is a definitional, qualitative study and a preprint.
- [E22] Chen, Zhang, Zhang and others (2025). Generative Interfaces for Language Models. https://arxiv.org/abs/2508.19227 (Preliminary). Does not support: That generated interfaces beat conversation in general. The abstract read here is a proposal and a preprint; its evaluation was not examined.
- [E16] Aljabri, Hijazi, Al-Salamin, Khemaja (2026). AI-MDD Approaches for Generating Adaptive and Intelligent UIs: A Systematic Literature Review. https://doi.org/10.20965/jaciii.2026.p1620 (Supported). Does not support: That adaptive generated UIs perform better for users. It maps the literature; the abstract read here does not report user outcomes.

## Limits

- The atlas is a broad, living catalogue, not every style that exists. No authority enumerates them, and the classes (movement, aesthetic, design system, layout, pattern, principle) overlap.
- The 1 to 5 scores are relative editorial judgements. They are not results of a study and not a measure of quality.
- An audience is a hypothesis about motivation, task and context. It was never measured and says nothing about any age, gender or other group.
- The sources support underlying principles (visual complexity, legibility, grouping, accessibility, platform conventions). None shows that a particular style works better or is preferred.
- The 2027 to 2030 trend is an editorial projection, not a forecast.
- Terms in the Future / AI category rest on exploratory work, some of it preprints, with no causal validation.

Last reviewed 2026-10-25.

## Heuristic scales

Relative editorial judgements from 1 to 5, not results of a study.

- Visual complexity: 3 / 5 (heuristic)
- Expressiveness: 4 / 5 (heuristic)
- Information density: 4 / 5 (heuristic)

## In our library

No design-style recipe implements this term yet; build it from `--ca-*` tokens.

## Prompt for your AI

```text
Design this interface in the "Generative UI" direction (future / ai, runtime composition): The UI is assembled or transformed for the task from context and intent.
Visual and interaction traits to use: components on demand, adaptation.
Main UX risk: Inconsistency, accessibility and trust. Mitigate it: Validated components, limits and inspection.
Keep it accessible to WCAG 2.2 AA: contrast, visible focus, keyboard operation, 24 px minimum targets, reflow at 320 CSS px and reduced motion.
Keep the content, hierarchy and behavior the same as a plain baseline so the two can be compared. Do not claim the style performs better; the choice should be validated with users.
```