The original iPhone felt magical because it made software feel physical, immediate and personal. For many of us, that was the moment we wanted to build for mobile.

Today AI has the potential to create that same kind of magic again but only if it feels native: fast, private, available, and close to the user.

In this talk, I’ll explore how on-device AI is becoming practical on Apple platforms: from Apple Intelligence and Foundation Models to Core ML, Core AI, Accelerate, and others. I’ll share lessons from deploying an actual on-device text-to-speech model in the ElevenReader app, including model packaging, performance optimization, latency, memory, battery, and model IP protection.

Just as personal computers moved computing away from mainframes, increasingly powerful devices and increasingly optimized models may move many AI experiences away from server-side inference. The cloud will not disappear but the most magical mobile AI experiences may be the ones that run right in your hand.