Cupertino frequently shows up late to events and acts as though it never happened. That’s about where Apple is now, two years after making a big announcement about AI features in 2024 that, for the most part, didn’t materialize on schedule. Though it’s still reasonable to question whether different equates to better, this year feels different.
Demonstrated at WWDC in June, the latest iteration of Siri is more than just an Apple-branded chatbot. It’s being designed to do things like draft emails, retrieve old hotel confirmations from a Gmail thread, and recommend items to bring to a friend’s potluck after you’ve read a text message. The last detail is what I find most memorable. An assistant is no longer truly an assistant if they recognize what you are looking at and respond accordingly. It’s more akin to having a silent partner right behind your back.
Apple refers to this as “personal context understanding” in conjunction with “onscreen awareness.” When the marketing jargon is removed, what’s left is a phone designed to understand you—your calendar, your messages, and your habits—enough to complete tasks before you’ve fully framed them as such. There’s a real benefit to that. There’s also something that should make people pause, even briefly, before handing over that level of access.
Private Cloud Compute, a system designed to ensure that requests handled off-device are neither stored nor visible to Apple or anybody else, is Apple’s response to the evident privacy concern. Outside researchers are permitted to confirm this claim. It’s a fair response, and Apple has gained some leeway in terms of privacy over time. However, “trust us, and you can check our work” is not the same as “we never touch your data,” and it’s important to distinguish between the two.
The person driving all of this under the hood is equally impressive. Notably, unlike OpenAI or Anthropic, Apple is not developing its own frontier model. Rather, it is relying on Google’s Gemini models for some aspects of the new Siri, with Anthropic supposedly having room to join later. For a company that has spent decades insisting it controls the entire stack, from software to silicon, that is a significant change. Apple appears to be wagering that orchestration—selecting the appropriate model for the appropriate task—is more important than model ownership. It’s a wise wager. Additionally, it’s a subtle acknowledgement that Apple fell behind on the one feature that consumers now demand from phones.
Hardware support has a narrative of its own. The more sophisticated features, such as sharper on-device dictation and expressive voices, are exclusive to the iPhone 17 Pro and the more recent Air model, while older iPhones receive the operating system update. For Apple, which has long used software as a lever to sell new hardware, that is not out of the ordinary. It does, however, imply that the “phone that thinks for you” experience won’t be shared equally, at least not initially.
All of this does not ensure that Apple does it correctly. The company has stumbled on Siri before, sometimes for years at a stretch, and the gap between a polished keynote demo and a phone that reliably understands what you actually meant can be wide. How frequently the new Siri will feel truly helpful as opposed to just competent is still unknown. It is evident that Apple is no longer making a covert effort to catch up. It is attempting to persuade a few hundred million iPhone users that it is worthwhile to let a machine think alongside them.
