This is a huge release! RuNNtime is webGPU backed inference engine for web with focus on performance. Perf is absurd: it’s close to 5x faster than best existing methods. Being able to run local models in real time with high fps opens so many new possibilities!

SwiftUI isn’t native and here’s why SwiftUI apps may sometimes feel janky. It ditches the key element that made animations run so smoothly on iOS since the original iPhone: Core Animation. In SwiftUI animations run in-app without CA backing. Block the main thread: SwiftUI ball stalls while UIKit’s keeps moving on the render server.

Here's what I think about the @ShopifyEng RN → native migration article: When I rebuilt the Expo Go UI using Jetpack Compose last year, I already felt that all declarative UI frameworks are fairly similar in use, and that agents do a good job migrating between them when it comes to presentation layers. So I'm not surprised Shopify found it possible to fairly quickly rewrite an existing RN app into two separate native codebases. That was always the easier part. I think the main problem lies in making sure the business logic layer doesn't diverge when there's nothing keeping it together. We introduced a few minor differences between the two Expo Go platforms despite trying very hard not to. And consistency vs platform feel is one more question of taste, not something an agent will always solve for you. I don't think react-native will go away, but I definitely feel react-native is evolving, will evolve further, and I have a couple of predictions: ➡️ More and more low level UIs as native components (either ExpoUI or local expo modules with native code files mixed with JS). Agents are great at writing presentation layers, so why bother having JS lists with all that complexity and perf costs? ➡️ OTAs are more important than ever, prepare for changes ahead of time. Slow app store review times vs cheap code push people to allow more customisability from JS, essentially treating the JS layer as server driven UI. ➡️ New UI paradigms mean RN remains valuable. Think custom, per-user interfaces (made for you UIs streamed inside chat, customizable apps etc.) – you need an interpreted higher level abstraction layer that's easier to define and has an SDK. ➡️ Focus on brownfield with no clear set boundaries. Agents pick and choose if they want to build experiences in native or in JS, with most components/APIs/libraries available both as JS and native entrypoints. In general: I don't think the falling cost of writing lower level code makes abstractions automatically obsolete. You get for free a lot of community effort on deciding/mapping platform quirks, stuff that would cost tokens and engineering effort otherwise, or even worse, would be missed and be a long-running paper-cut in your app. I think it makes it necessary to make abstractions lock-in free. Use these opinions and higher level components when they make sense and give value, use lower level super easily and cheaply otherwise. That's why I'm excited about stuff like Expo UI, the new modules API, our @swmansion libraries that work for both RN and native, and all the other effort we did as a community to make mixing those approaches possible and pain free.

Got a ticket for agent conf in Warsaw on Sept 17-18 courtesy to @grabbou Unfortunately can’t come so will give my ticket to whoever edits the below photo to give me the biggest “dziubek” (duck face) possible. The contest ends in 24h

So much to explore at the intersection of real-time computer vision and graphics (two areas where GPUs excel). Agents are really good at writing TypeGPU. Install TypeGPU skills, point your agent at examples from the repo, and build something cool!