Generative UI
Interfaces assembled at runtime by a model, rather than laid out in advance by a designer.
Design DisciplinesIn generative UI, the model does not return text — it selects and composes interface elements for the specific request. Ask for a comparison and get a table; ask for a trend and get a chart; ask to book something and get a form with the known fields filled. The designer's work shifts from drawing screens to defining a constrained vocabulary of components the model may assemble, plus the rules that govern the composition.
The hard part is that consistency and learnability are properties of repetition, and generated interfaces break repetition by design. If the same question produces a different layout each time, nobody builds a mental model of the product. The practical approach constrains hard: a small component set, fixed layouts per intent, generation limited to which components appear and what data fills them — not to inventing new arrangements.
It is early, and the failure modes are unsolved
Generated interfaces make testing, accessibility and support genuinely harder — you cannot audit a screen that does not exist until runtime, and a support agent cannot see what the user saw. Treat these as open problems rather than details to sort out later.
In practice
An analytics product let a model choose any chart type per query. The same question asked twice produced a bar chart and then a line chart, and users stopped trusting the results. Restricting generation to a mapping from query shape to one chart type — with a manual override — kept the flexibility people wanted and removed the inconsistency they did not.
Where teams get it wrong
- Unconstrained generation, so the same intent produces different layouts and nobody learns the product.
- No accessibility guarantees, because generated compositions are never audited.
- No way to reproduce what a user saw, which makes support impossible.
- Generating layout where only content selection was needed.
- Using it where a designed screen would be better, because the technology is available.
Learn more
You may ask
Frequently Asked Questions
What is generative UI?
An interface assembled at runtime by a model, which selects and composes components for the specific request rather than rendering a layout a designer drew in advance.
What are the risks of generative UI?
Inconsistency that prevents people learning the product, compositions that were never accessibility-tested, and no reproducible record of what a user saw — which makes support and debugging much harder.
Related terms
All terms- AI Product DesignDesigning products whose behaviour is probabilistic — where the system can be confidently wrong and the interface has to account for it.
- Prompt DesignDesigning what a product asks a model, and what it asks the user — the interface layer around a prompt, not just the wording of one.
- Design SystemThe single source of truth for how a product looks and behaves: tokens, components, patterns, usage rules and the process for changing them.
- Component LibraryA shared, versioned set of reusable interface components that designers and engineers build from instead of recreating elements per screen.
Defined by Mara Last reviewed .
Let's talk about your product.
Happy to look at what you're building and say where design would move the needle.
Contact Us