The True Test of Artificial Intelligence
Forget AGI; I want something better. When AI can correctly implement an Apple TV interface, with every gesture working in the focus engine, humanity will be ready for the stars. Claude and Codex flounder; Grok eventually got there. SwiftUI plus tvOS remains a shockingly awful combo in 2026.
I realize this is moving goalposts and all. I'm not interested in humanity achieving AGI; we need something better. When AI can correctly implement an Apple TV user interface, including making all of the gestures work correctly in conjunction with the focus engine – now that will be an achievement. Monumental. Unbelievable. Humanity will be ready to explore the stars. That is to say, I wouldn't count on us achieving this anytime soon.
I, personally, was able to achieve relative peace with the tvOS focus engine back when I was working with UIKit. Once I introduced SwiftUI into the mix, everything went downhill. How naive was I in thinking, "Oh, this'll be great: a fresh project, starting with all of the latest tools!" All the latest failures, is more accurate. SwiftUI remains half-baked in 2026 and the tvOS focus engine remains unimpressive and ill-suited for the task at hand.
If you've never implemented a user interface in tvOS, count yourself blessed.
I've worked extensively with Claude (Opus) and Codex (GPT) on tvOS focus engine issues (post UIKit). They both flounder. I pulled in Grok Build more recently and the first thing it did was drop the built-in component and proceed immediately (without even asking) with writing a replacement component from scratch. That ended up being the right way to go, but it was an hours-long endeavor. I had previously attempted that with Claude, and it couldn't get the tab bar to function at all. SwiftUI + tvOS is a shockingly awful combination.
