Just-in-Time UI
Just-in-Time UI refers to user interfaces that are generated by an AI system only at the moment of use, tailored to the specific task at hand. Instead of fixed menus and forms, exactly the view that is needed at that moment is created each time.
Every app has a user interface: buttons, input fields, menus, lists. This interface was designed by humans beforehand and built permanently into the program. It looks the same for all users, regardless of what they want to do. With Just-in-Time UI, this is different: the interface is only assembled at the moment you need it. An AI system understands your request and generates matching buttons, sliders, or tables for it. For the next request, the interface looks different again. “Just in Time” comes from industry, where parts are only delivered when they’re needed on the assembly line.
What chat windows can’t do
Text chats with AI systems are practical, but inconvenient for some tasks. If you want to pick the best flight out of twelve options, a text list is cumbersome. A table with sorting buttons would be much quicker to grasp. This is exactly the gap Just-in-Time UI is meant to close: language for the intent, visible controls for the fine work.
The second reason is the overload of classic software. An image editing program has hundreds of functions, but you rarely need more than five of them. Developers have to anticipate all possible cases in advance and fit them into menus. A generated interface only shows what belongs to the current task and leaves out the rest.
For companies, this is also a cost question. Programming special views for small groups of users often isn’t worth it. If the AI generates such views on demand, this threshold drops significantly. That’s why Just-in-Time interfaces show up in business news as an argument against the value of classic software vendors.
From sentence to control element
Technically, this is usually powered by a language model, i.e., an AI system that understands and generates text. But it doesn’t output images—it outputs code or a structured description. This description states, for example: here a slider from 0 to 100, next to it a list with three columns. The app reads this instruction and builds the visible interface from it.
To prevent anything wild from emerging, developers work with fixed building blocks. The AI may only choose from a catalog of prefabricated elements and arrange them. So it doesn’t reinvent every button, but combines known parts. This keeps the result stable, usable, and visually consistent.
A common misconception is that entire programs are created here. Usually, only the interface is generated, while the actual work continues to be handled by fixed programs or databases. So the AI connects intent and existing functions. One risk remains: if the view changes every time it’s called up, you can’t memorize any routines. Good implementations therefore deliberately keep recurring elements in the same place.
First examples in real products
Today, the idea is most visible in chat assistants from major providers. If you ask about the weather, a small weather card appears instead of a paragraph. In purchase advice, a comparison grid with filters sometimes appears. These insertions are early, still limited forms of Just-in-Time UI.
This is also being tested in enterprise software. A sales rep says which customer data they want to see and receives a matching analysis with sliders and charts. Previously, they would have had to use a standard dashboard for this or ask the IT department.
In news coverage, the term often appears together with “agents”, i.e., AI systems that carry out tasks independently in multiple steps. Such agents need a point where the human checks and confirms. A generated interface is exactly that point. Anyone wanting to follow the development should watch whether providers are dismantling fixed menus or only offering generated views as a supplement.