Generative UI and early thoughts

Today’s interfaces are deterministic. Designers and engineers define exactly what appears on the screen, in what order, and under which conditions. If the product needs to behave differently for different users or situations, someone must manually redesign or reimplement that behavior.

Generative UI explores a different model. Instead of hard-coding the interface structure, we describe the intent and the design grammar, and let an LLM generate the most relevant structure at runtime using real-time context. The output is probabilistic rather than fixed, which makes the UI more adaptable without new feature work.

Below is a simple example that contrasts a deterministic set of tabs that never changes with a generative set of tabs that is produced dynamically based on the user’s query and context.

This approach opens up new possibilities for product design. Instead of predicting every scenario upfront, we can let the system adapt in real-time based on user needs and context. It’s a fundamental shift in how we think about building interfaces.