Runway has just thrown a grenade at the idea of the fixed app. Solaris, the first model in what the company calls “Interface World Models,” doesn’t run software as code. It generates the interface frame by frame in 720p, responding directly to clicks, drags, and voice commands. No translation from design to code. No waiting for an update. The model is the app.
The use cases Runway floats are genuinely interesting. An online store that rebuilds itself for each shopper — drag a shirt onto your photo to try it on. A furniture configurator where saying “move the table” actually moves the table. Tutorials that show you the next step in context instead of spitting out a wall of text. These aren’t incremental improvements. They’re a different mental model for what software even is.
Solaris builds on Gen-4.5, Runway’s video model, with a language model deciding how the interface should change while the world model renders each frame. The company also frames this as training infrastructure for AI agents, which still fumble real-world tasks like hotel bookings because they’re trained on fixed layouts that break the moment a website deviates from the script. An agent that can generate its own interface on the fly wouldn’t have that problem.
But the gaps are real. Runway admits stable, readable text is still hard, which is a problem when interfaces are mostly text. Long sessions and reliability are open questions. And a generated interface that looks right but behaves wrong is arguably worse than no interface at all. Screen reader support? Not there yet. Runway is calling this a research effort and looking for launch partners through an early access form. The company’s larger bet — that the app as a fixed unit disappears entirely — is bold. Whether users actually want interfaces that keep changing is a question Runway leaves pointedly unanswered.
