Most on-device AI talks stop at wrapping someone else's model. The real work starts when the model you need doesn't exist yet. You have to adapt it, get it running on a phone, and ship it behind a clean API. And the steps nobody likes to talk about are the ones where projects quietly die: data prep, and export.
Here's the thing. Those steps are fiddly, repetitive, version-sensitive, and unglamorous, which makes them exactly what coding agents are good at.
This talk follows one custom model the whole way: adapted for a real task, exported to run fully on-device, and shipped as a React Native hook you can actually call. At every hard step, I hand the grind to an agent and guide it through the whole process.
You'll walk away knowing how to ship a custom on-device AI feature in React Native, how to direct an agent through the parts that usually break, and how to get it to do the work you've been dreading.