The problem
The first version was a fairly standard dropdown. Functional, but limited once the number of models and media types started expanding. The output type — text, image, or video — was effectively invisible.
The approach
I designed a new selector with simple and advanced modes: an autorouter for users who just want an answer, and manual selection for those who want more control. Models are grouped by capability, and the active output type is always visible.
Case study in progress — more detail and process work coming soon.