▶ YouTube Summaries

• YT :: https://www.youtube.com/watch?v=zqOrriq20Tc
• Original title :: Apple Changed...

A long rant, triggered by a 36-hour wait for Apple to approve installing his own app on his own iPad, arguing that iOS has regressed from Steve Jobs' original "no SDK needed, just ship a web app" promise into a platform whose sandboxing, App Store rules and Xcode toolchain actively prevent experimentation — and that this now caps what AI-era software can become, on Android too, since Android mostly copies iOS.

Framework: importance, cost, capability

Software can be rated on three independent axes — how important it is, how expensive it is to build, how capable it is. The iPhone raised importance and capability enormously while raising cost moderately, a fair trade. Since the App Store launched, capability of everyday mobile devices has gone down year over year while cost went up.

Jobs' 2007 pitch (quoted from the keynote) was: full Safari engine, apps integrate with phone/mail/maps, instant distribution from your own server, easy updates, no SDK. Theo's point is that the current state is a regression from what was promised before the App Store existed.

Why Apple stopped caring

Apple takes 30% of app purchases and of every digital good sold inside apps. Globally Apple has <20% device share but >60% of money spent on mobile; a company with $100M/year mobile revenue typically sees 60–70% of it on iOS. So Apple never has to compete on developer experience — developers must support iOS regardless. During Epic v. Apple it came out that Apple makes more on games than Sony, Microsoft, Nintendo and Activision combined, without ever shipping a game.

Signing and provisioning hell

• Connecting a new iPad to his own Mac left provisioning "pending" for 36 hours; he could not install software he paid to build onto hardware he owns.
• TestFlight is no escape — it still goes through full app review.
• Devices must reconnect to the signing machine every 7 days or the app stops opening.
• Signing credentials live hardcoded in Info.plist, so a developer-account-specific chunk of a shared file gets checked into git.
• After approval he still could not build: an obscure error pointed at a new post-WWDC agreement hidden behind a banner on developer.apple.com, plus another 20–30 minutes of server-side propagation.
• You cannot build an iOS app without a Mac (or someone else's Mac in the cloud).

The whole regime is hostile to experimentation and to open source: forking somebody's approved app means redoing the entire signing dance and dirtying files you must not commit.

Sandboxing kills whole categories

Apps cannot talk to each other without an external server. You cannot customize apps the way a Chrome extension customizes a website, cannot force dark mode, cannot have an org-wide app tweak. JIT and any dynamic code execution are banned, so performant emulators are impossible even though emulators are nominally allowed; even React Native sits near the line. Only WebKit is permitted as a browser engine. Plugin/extension ecosystems inside third-party apps effectively cannot exist.

App Store policy

The only routes onto a phone are the App Store and TestFlight, both gated by Apple review.

• Streaming video and music: fine. Streaming games: banned.
• Any third-party OAuth forces you to also offer Sign in with Apple, even at 0% usage.
• External payments are banned: currency bought on desktop must also be purchasable in-app through Apple Pay at 30%, or you are banned.
• Physical goods and in-person services (Uber, DoorDash, TaskRabbit) pay 0%. So the rate is 30% or nothing depending on category.
• The Apple Video Partner Program (15% instead of 30%) exists because Amazon threatened to pull Apple hardware from Amazon.com; Amazon is effectively the only member.

The 30% destroys business models built on thin margins: Patreon (5–15% take), Twitch subs, any GoFundMe-style app taking only Stripe fees. At Twitch they had to invent a currency — buy $6.50 of credit (Apple takes its cut there), then spend it on subs — to comply.

He also calls out review arbitrariness (three separate rejections because the reviewer typo'd the test login email) and, added as an afterthought, that scam weekly subscriptions aimed at children are a revenue line Apple declines to police — his claim: no weekly subscription in the App Store is not a scam.

Xcode

Claim, sourced to Apple employees: the core of Xcode is still the NeXTSTEP codebase Apple bought from Jobs, over 25 years old. The symptoms he lists:

• incremental builds slower than clean builds; no-op changes trigger multi-minute rebuilds; long opaque phases like "planning Swift module"
• nondeterminism that makes "clean build folder" and "delete derived data" folk medicine; errors that vanish on restart and jump to random files
• Swift's type checker giving up ("unable to type check this expression in reasonable time"), forcing refactors for the compiler's sake, worse under SwiftUI's ViewBuilder
• archive builds behaving differently from run builds
• no module system, hence third-party Ruby CocoaPods — whose author left to contribute to TypeScript at Microsoft
• SwiftUI previews nobody uses; LLDB unreliable; simulators needing reinstall, erase-all-content, huge runtime downloads, disk bloat, and device/simulator behaviour mismatches; Xcode 26 broke clipboard paste into the simulator
• project.pbxproj: constant merge conflicts, unreadable diffs, IDE mutating project state, unopenable project if malformed — a pre-git worldview that is being ignored rather than fixed

His hottest take: Xcode can never be good because the Xcode team has only ever built Xcode with Xcode on Apple platforms — they have no reference for what good tooling feels like. And he pushes back on the "you get used to the quirks" attitude from iOS devs as Stockholm syndrome.

The AI argument

The best AI experiences are not features inside apps — they are layers that change how you use the whole machine: a push-to-talk dictation overlay (Whisper Flow), Codex driving the browser in another window or using the computer with the screen off, pulling context across apps. None of that is expressible on iOS, where Siri is the only layer with OS privileges and third parties can at best be called by it. WhisperFlow on iOS has to ship a keyboard that bounces you into its own app and then leaves the mic on via a live activity. So we will never learn what phone AI could have been.

Android does not save us

Quoting Scott Jensen (ex-Google/Android, ex-Canonical): a VP told him "the fastest way to ship your features is to have Apple do it first." Two-thirds of these problems exist on Android not because Google is greedy — Google makes little on Play — but because Google copies Apple: the 30% cut, the review process, the OS-level restrictions. Android is B-tier iOS by choice. If Google treated Android like Chrome instead (e.g. embedding something like Expo, allowing user-built overlays), the world would look different. Theo is toying with his own Android fork, not expecting it to go anywhere.

What could have been

A 10-year-old building an iPad app on their iPad and sharing it with friends. Customizing a device around a disability. Choosing Claude or ChatGPT or Gemini as the OS assistant with real primitives. Keyboards not capped at ~50MB. The no-SDK world Jobs described in 2007.

His closing plea is aimed at Apple's new CEO John Ternus: let the hardware win, let software experiment. The damage is generational — kids grow up never learning software is malleable, and only meet a general-purpose computer in their late teens. Sign-off: buy a kid a Linux computer.