Beyond who an AI interface is for and how long it lasts, what decides how stable it stays and who's steering?
What does generative UI design space split into beyond audience and timeframe?
This explores the other two dimensions of the generative UI design space, constancy and control, and what each one means for interfaces that AI builds on the fly.
This explores the two dimensions of generative UI that sit alongside audience (who the interface is for) and timeframe (how long it lasts): constancy and control. A CHI workshop proposal lays out all four together How should generative UI be organized as a design space?. Constancy asks how stable an AI-generated interface stays. Does it keep the same layout from one session to the next, or rebuild itself each time? Control asks who steers the result: the user, the system, or some mix of the two. The framework comes out of older research on adaptive and malleable software, which kept running into the same tension. When a system adapts too much, users lose their bearings. When it adapts too little, it isn't really generative.
The control dimension gets more interesting once you ask who decides that an interface should exist in the first place. Moran draws a line between generative UI, where the AI decides what to build, and vibe coding, where the user asks for something specific Does who initiates a UI element matter for accountability?. That changes what counts as failure. If the AI chose the interface, a bad result is a failure of design judgment. If the user specified it, a bad result is a failure to carry out instructions. Moran also argues that handing control to the AI is the whole point for many people, because most users lack the interface vocabulary to describe what they want Can generative UI serve users who cannot specify interfaces?.
Giving the AI that control only helps if the AI follows through. The Design Theater benchmark found that generative UI tools skip about a quarter of the design rationales they state. For functional requirements the figure rises to 34%, and the tools recognized only about half the UX principles written into their prompts Do generative UI tools actually implement their stated design rationales?. So the gap between the user and the system has a second layer: a gap between what the system says it is doing and what it actually builds.
For wider context, a review of 291 papers on human-GenAI interaction found that user control over the model is one of six recurring design dimensions How should designers structure human-GenAI interaction systems?. Control isn't unique to generative UI. It shows up across generative AI design in general. AWS's Cloudscape design system turns it into a concrete rule: always let users override the AI How should generative AI design embed responsible AI principles?. One takeaway: constancy and control are linked. The more an interface reshapes itself, the more users need a way to take back the controls.
Sources 6 notes
A CHI workshop proposal identifies audience, constancy, control, and timeframe as the four dimensions structuring generative UI design. The framework emerges from reviewing prior adaptive and malleable-software research, mapping tensions between user control and system adaptation onto a unified design space for AI-generated interfaces.
Moran argues generative UI (AI-initiated) holds systems accountable for design judgment, while vibe coding (user-initiated) holds them accountable for execution fidelity. This origin point determines what failure means for each mode.
Moran argues genUI's advantage is that AI acts as designer, deciding what interface to build rather than requiring users to specify one—addressing the gap where most people lack the interface literacy that vibe coding presupposes.
A benchmark of 24 tasks across five tools found roughly 25% of design rationales go unimplemented, rising to 34% for functional requirements. Tools recognized only half the UX principles embedded in prompts.
A systematic review of 291 papers identified six recurring dimensions: purposes for using GenAI, model feedback to users, user control over models, engagement levels, application domains, and evaluation strategies. Three authors collaboratively coded and synthesized these through iterative consensus.
Show all 6 sources
AWS's design system translates fairness, explainability, privacy, robustness, and governance into five interface rules: evaluate use cases before automating, disclose AI generation and low confidence, obtain consent before using customer data, test continuously for bias, and enable user override. These sit upstream of interaction design rather than addressing stochastic output variability.
Papers this line draws on 8
The research behind the notes this line reads — ranked by how closely each paper relates.
- What does Generative UI mean for HCI Practice?
- Design Theater: Evaluating the Gap Between User-Facing Design Reasoning and Implementation in Generative UI Tools
- Prototyping with Prompts: Emerging Approaches and Challenges in Generative AI Design for Collaborative Software Teams
- Evaluating Generative AI in the Lab: Methodological Challenges and Guidelines
- Design Principles for Generative AI Applications
- Generative UI: LLMs are Effective UI Generators
- Large Language Models for User Interest Journeys
- GenUI vs. Vibe Coding: Who's Designing?