Generative UI

What

Interfaces that are not drawn in advance but assembled as they run. Where the model decides, and where the rules a person wrote must win — tried inside real products.

Why

Let a model build the whole screen and consistency and trust go first. Fix everything in advance and there is no reason to use a model at all. The usable answer sits between: the content from the model, the form from the system.

Process

• The model proposes, a person decides – SAI – the AI offers what it heard that might be useful, and only what someone accepts becomes a note. The suggestion card is a fixed component with four states (offered, working, done, dismissed); only what is inside it is generated. Confidence shows as the fill of the card, and what it doesn't know it says it doesn't know.
• Never without evidence – SAI – every suggestion carries the utterance it came from. Nothing appears without one.
• Summaries from the model, facts from the data – Commit – the AI summarises a notice; the budget, dates and eligibility come from normalised data. Runtime model calls were removed over cost and hallucination.
• The model's probabilities as a picture – Distance of Thinking – next-token probability drawn as a space, so what the model is weighing can be seen.

Result

• Fixed form, generated content. A component has to stay put for what someone learned to keep working.
• Generated text without evidence never reaches the screen. Accepting a suggestion means seeing what it came from.
• Runtime generation is a cost. Whatever could be computed in advance was.