John Gruber's Apple Intelligence post-mortem: 'That's All She Wrote'
Mar 13, 2025
Key Points
- Apple delayed personalized Siri features from 2024 to 2026 after announcing them at WWDC, signaling execution failure at a trillion-dollar company that should have shipped working products.
- Apple demonstrated only concept videos for its most ambitious AI features rather than working prototypes, a red flag that hasn't appeared since 1987 when the company was near bankruptcy.
- Siri's core dictation still fails on basic tasks like technical terms and product names, undermining any agent feature Apple builds since users won't trust voice input that misunderstands commands.
Summary
John Gruber published a post-mortem on Apple Intelligence that reads as an indictment of execution failure at a trillion-dollar company. Apple announced personalized Siri features scheduled between now and WWDC, then quietly delayed them to 2026. Gruber says he missed the red flags and is embarrassed about it.
The real story sits in what Apple showed versus what it didn't. At WWDC 2024, every Apple Intelligence feature that has actually shipped to users was demonstrated in controlled conditions by Apple reps. The personalized Siri features—the ones requiring awareness of personal context and cross-app action—were never demoed. Apple showed them as a concept video instead.
Concept videos are a warning sign. They signal a company in trouble. Modern Apple, post-Jobs and post-Next reunification, does not publish concept videos about future products. It demonstrates working products. The last time Apple leaned on concept videos was 1987, when the company was heading toward near-bankruptcy. Gruber's framework should have been straightforward: some features are table stakes like text-to-speech and speech-to-text. Others are ambitious and risky. Apple only showed the commodities.
The practical failure compounds the strategic one. Gruber tested Apple's Siri dictation on his iPhone and it couldn't handle basic commands. It couldn't transcribe technical terms, footnotes, or product names. This matters because any agent feature Apple builds depends entirely on trust in voice input. If Siri misunderstands "Palace of Fine Arts" and routes you to Oakland instead, you lose money and trust simultaneously. Apple hasn't solved the fundamentals.
Apple has every advantage to own the consumer AI agent market. It has direct control over device APIs, access to personal data, the most money, and the best talent. Yet it's shipping mediocrity while startups race to define what a useful agent even is. Nobody can articulate what they actually want from a phone agent yet. That's the opening for YC founders to discover something Apple, as a trillion-dollar enterprise, may not.
Gruber closes with the MobileMe precedent. In 2008, Apple launched MobileMe as a broken product. Steve Jobs held an all-hands meeting and asked what the product was supposed to do, then asked why it didn't do that. He called out the team for tarnishing Apple's reputation and installed new leadership on the spot. Gruber's inference is sharp: Tim Cook should have already held that meeting for Apple Intelligence. If he hasn't, and doesn't soon, the company has lost its way. Gruber writes this out of love and disillusionment, having spent two decades defending Apple.