Chris Tate tests Jev with json-render to assemble app interfaces at runtime

BlockBeats reported that Vercel Labs engineer Chris Tate has connected Jev to json-render and begun testing a runtime approach to UI assembly.

Under the setup described in the report, json-render prepares buttons, cards, input fields, data, and actions ahead of time. Jev does not generate a full JSON structure token by token the way a large model would. Instead, it decides which components to use, how to order them, and where to place them, then passes the result to a standard UI renderer.

Demo cut generation time to under one second

In the demo setup, the system first makes one decision to pick the main components for the interface. If needed, it then makes a second decision to arrange the layout.

For the same train-ticket interface generation demo, the default JSONL mode took 3.21 seconds from request to completion, while Jev took 0.88 seconds. The report said the full generation time in this demonstration was pushed below one second, which gave this kind of interaction a real-time feel.

App interfaces may stop being fully fixed in advance

According to the report, future app interfaces may not need to be completely predefined. UI could become a runtime decision, allowing the same app to assemble different pages on the fly for different users, tasks, and states.

How Jev differs from Runway's Solaris

BlockBeats also noted that Solaris, released by Runway late last month, is aimed at a similar direction but takes a more aggressive route. Solaris uses a world model to generate the entire interface frame by frame, bypassing traditional code and components. Jev still uses components and actions prepared by developers in advance, while dynamically deciding how to combine them.

The report framed the contrast this way: one path wants software to become a real-time generated video world, while the other makes software look more like something with a face that can be reorganized at any moment.