What happens when UI stops being passive?

UI is evolving.

For years, it was deterministic. You clicked a button, something happened. You selected an option, the system responded in a predefined way.

Then LLMs came, and UI became more generative. Instead of telling a system exactly what to do, you could describe what you wanted and it would generate the output, and increasingly the interface for you too.

But I feel there's another evolution coming: proactive UI.

Speaking to a few people in the industry, and observing how some of these products are evolving, it's evident that the system is starting to move beyond simply responding to what you ask. It's beginning to understand more about what you're trying to achieve, what it should know, and when it might be useful to contribute. We're already starting to see patterns of this emerge in tools like Claude Code and Cursor with planning, but I feel there's a layer beyond that.

How might this change the way we design interfaces?

First, it starts helping construct the input. Rather than simply asking the user to provide a prompt, it can recognise what context is missing and help them provide it. A Figma file, design system, existing code, documentation, examples, constraints. In other words the interface starts helping the user understand what information will make the outcome better. Claude Design does this in part. Before the system builds anything users go through a series of questions to narrow down their prompt, creating a more tailored and relevant output.

Second, the system has to have somewhere to surface things. If it can identify something useful without being asked, that information has to have a place to go. Perhaps it appears in the workflow, as a suggestion, a card, a notification or something more ambient. We're no longer just designing where the user goes to interact with the system, but where the system can come to the user.

Third, timing becomes part of the UI. A useful suggestion at the wrong moment is not useful. The system has to consider not just what to surface, but when. The interface becomes temporal.

Fourth, there has to be a hierarchy of intervention. There's a big difference between "you might want to look at this" and "I've changed this for you". Proactive systems have to distinguish between when to suggest, when to ask, and when to simply act.

And finally, feedback becomes part of the interaction. Opening something, ignoring it, editing it, dismissing it or repeatedly asking for something different all become signals. The system can start learning not just what you want, but how you operate and what's of value.

Not every interaction will require this. Sometimes you just want to click a button for an immediate action to take place.

But I feel the interesting challenge for the next generation of interfaces is no longer simply how do we help the user interact with the system?

It's how do we design a system that knows when to and how to contribute effectively to its user?