Meet Core AI

A Podkey summary of Meet Core AI, from WWDC 2026.Today’s catch-up is really about what happens when AI moves off the cloud and onto the device in your hand. The big theme is cost, speed, and privacy all getting reshuffled at once, with Apple’s Core AI approach making local inference practical, while developers get better tools for startup time, debugging, and performance tuning. And underneath the nice demo version of that story, there’s a lot of very real engineering around transformer caches, compilation delays, memory safety, and custom GPU kernels.Why on-device inference changes the economicsThe transformer latency problem and the cache fixWhy startup delays matter more than people thinkAhead-of-time compilation helps cut the waitA Swift API that stays safe without getting in the wayWhen generic kernels aren’t enoughDebugging the math and spotting bottlenecksWhy this all fits togetherThis podcast was created with Podkey. Make your own at https://podkey.fm