Apple Just Won AI (and It’s Not Even Close) (Video)

Apple may have just won consumer AI by being last, and here is the twist that makes it interesting: it did it without having the smartest model. ChatGPT is more powerful. Claude is better at complex reasoning. Apple even leaned on the tech behind Google's Gemini to help train its own models, and after promising a revolutionary Siri back in 2024 and then failing to ship most of it, Apple had every reason to be the company nobody trusted with AI. But while everyone was writing them off, they quietly rebuilt the Siri that had been broken for 15 years, and after spending time with the beta, I think that changes the whole equation.

Here's the thing I keep coming back to: Siri's real problem was never that it was dumb. It was dumb, don't get me wrong, but mostly it was trapped. Since it launched on the iPhone 4S in 2011, Siri could set a timer or call mom, but it couldn't see the app in front of you, understand that "send him the address" meant the person in your current conversation, or connect an email to a calendar event to a text message and hand you one useful answer.

The information was right there on your phone. Siri just wasn't allowed to reach it. At WWDC 2026, Apple showed the architecture built to tear those walls down, using five custom models and an orchestrator baked into the OS that decides which model handles your request and whether it runs on-device or in the cloud. 

The distinction I want people to sit with is intelligence versus access, because they are two completely different advantages. Ask Siri what everyone's bringing to the neighborhood potluck and it can search your messages, email, and calendar and just tell you. Ask ChatGPT the same thing and it has no idea because it can't see the group chat, and doesn't know which Saturday you mean.

That's not a knock on ChatGPT, which is genuinely brilliant at hard problems. It's that using a third-party model for something personal on your phone still starts with homework: open the app, describe the situation, go dig up the email or screenshot, come back, paste it in. Apple starts from a different place. It owns the OS, the chip, the secure enclave, and the privacy framework, so a lot of that context can be handled right on the device, with Private Cloud Compute extending it when a bigger model is needed.

That's why the side button matters so much, and why the demo of Mike Rockwell just holding it down and asking about concert tickets felt so ordinary. The ordinariness is the breakthrough. Winning consumer tech disappears into behavior.

I did run the competition. OpenAI and Anthropic have the models but not the phone. Google has nearly everything but Android is fragmented across hundreds of manufacturers and Pixel is a tiny slice. Microsoft owns the desktop but not your pocket, Meta's glasses are impressive but still ride on a phone it doesn't control, and Amazon never got the smartphone at all.

Everyone's fighting from a structurally weaker spot. Add visual intelligence, Siri seeing through the camera and eventually through glasses, and Apple's runway gets longer.

Now, nobody should crown Apple off a keynote, and I say that as someone who watched them do exactly this in 2024 and then pay $250 million to settle false advertising claims over Siri features that never shipped. The new Siri is beta, English-only, and Apple still has to prove it can keep distilling frontier advances into its own models without rebuilding everything each time Gemini leaps.

My bet, and Apple's bet, is that good-enough intelligence with exceptional context beats exceptional intelligence starved of context for the everyday 90 percent. Siri handles the frequency, the frontier models handle the complexity. In my testing the beta has been superb, and if it holds, the most important assistant won't be the one topping benchmarks. It'll be the one you can ask about your life without explaining your life first.