There are three things I want to cover in this post: the Snow Leopard rumor, the Siri shenanigans, and Private Cloud Compute. Beginning with the Snow Leopard rumor, here’s Mark Gurman, reporting for Bloomberg back in February:

Apple’s iOS 27 cleanup job includes code changes that could improve battery life. When it comes to Apple’s operating system updates this fall, the company has two main priorities. First, it wants to integrate AI into the software with a revamped, chatbot-style Siri. Second, the company is looking to clean up an operating system that has become a bit of a mess under the hood.

The tidying effort (similar to what Apple did with its Snow Leopard OS) will include removing scraps of old code, rewriting features, and subtly upgrading apps to let them perform better. That should make for a snappier, more responsive operating system. Apple also is planning some tweaks to the interface, though nothing as extensive as last year’s Liquid Glass introduction.

As part of the work on iOS 27 — code-named Rave within Apple — engineers are making improvements that could boost battery life, I’m told. The company hopes that underlying code changes will result in efficiency gains that will end up giving users more juice. If Apple does indeed pull off these improvements, it’s unclear if the company will market the changes — or if this is just a benefit of cleaning things up.

I think “Snow Leopard” is actually quite an apt description for this year’s operating systems. Mac OS X Snow Leopard was famously described as having “Zero New Features” onstage at 2009’s Worldwide Developers Conference. Indeed, the software had few customer-facing changes, with Apple instead opting to give the OS much-needed performance and stability improvements. Part of the underlying cause of these improvements, however, was an internal feature called Grand Central Dispatch, which introduced multithreading to OS X for the first time. Apple apps — and third-party apps later in the fall — were updated to run multiple processes concurrently, taking advantage of new Macs’ multiple cores.

Grand Central Dispatch is so important to modern iOS and macOS that we don’t even use the name anymore. Any processor-intensive task on iOS and macOS uses multithreading. (Just check Activity Monitor → Window → CPU History.) It was one of the most important features of OS X and set a precedent for how operating systems would work in the coming decades. Similarly, I imagine the new “more personalized Siri” and focus on on-device artificial intelligence — if these features are released into beta next week — will soon become integral to the way we use Apple devices. That was the thesis of my 2024 post-WWDC article “Why Apple Intelligence Is the Future of Apple Platforms.” On-device intelligence can repair AI’s suffering public image and takes full advantage of the powerful Neural Engines found in the latest Apple silicon.

But at its core, Apple’s platforms desperately crave some attention to detail. They’ve been deprived of engineering talent and design intuition for years, and it is starting to show. With the help of AI agents, I hope Apple revisits the SwiftUI apps written at the beginning of the decade and refines them with more capable UIKit and AppKit components. (System Settings, especially.) Conversely, I know there is plenty of old Objective-C code that could do with a modern rewrite — Swift is vastly more efficient than Objective-C, and with younger Apple engineers coming only with experience in Swift, now seems like a great time to begin the transition to newer components slowly. I understand the apprehension from some seasoned developers, but this transition must happen at some point. Whether it should include SwiftUI or not in the agentic age is debatable.

I also hope Apple’s new leadership lends greater importance to user experience. Alan Dye, Apple’s ex-software designer who left for Meta in December, famously neglected the Mac and its idiomatic controls and axioms. I hope Steve Lamay, Apple’s new software designer, pays greater attention to finer details, both on iOS and macOS. I hope new leadership, plus the unwavering spirit of AI agents, will have an important impact on the quality of Apple software come next week.

Bloomberg’s Gurman also has been extensively reporting on the “more personalized Siri.” Here’s him in May:

Apple has redesigned Siri for modern iPhone hardware, making it live inside the Dynamic Island as an always-on agent that can help users get things done across the operating system and within apps. The system can draw on web data, personal information, and what’s on a user’s screen to complete tasks.

There are two starting points for Siri. The classic approach — saying “Siri” or holding down the iPhone’s power button — is set to trigger a redesigned Siri animation in the Dynamic Island, the pill-shaped screen element that Apple introduced in 2022. That mode is best suited for voice-based queries and search.

The second method is entirely new: Apple plans to let users swipe down from the top center of the iPhone anywhere in the system to launch a new Search or Ask interface. (The Notification Center can now be opened by swiping down from the top left.) That opens a revamped Siri experience designed for getting things done or searching by typing, though using voice remains an option.

I’m quite intrigued to see how this interface looks in the final version of the OS, because Gurman’s reporting seems incoherent to me. Is the interface that appears when holding down the Side Button the same as the one hidden behind the Dynamic Island? What about iPhones without the Dynamic Island? What about the Mac and iPad? What differentiates the Siri mode from the Search and Ask interface? What is the difference between searching and asking? I think the answers to these questions will be a surprise come next week. Broadly, I think Apple has solved the three modalities: web search, device search, and agentic actions on the web. How it will integrate those three modalities in this interface is Monday’s question.

If I had to guess, I’d say the Side Button immediately launches the voice mode, while the Dynamic Island swipe gesture mirrors the current “Talk to Siri.” (I have Talk to Siri disabled because I find I activate it incidentally with the current double-tap-the-bottom gesture. This redesign will alleviate that pain.) “Ask” and “Siri” are different modalities for the Siri chatbot — text and voice, respectively — powered by on-device Apple Intelligence models and Gemini models in the cloud. The Ask feature will use a web search tool at its discretion. This is how I’d design the interface, at least. App Intents or Model Context Protocol integrations will likely be built into the Ask interface as tools, as will the Personal Context feature unveiled in 2024.

I’m also anxious to see how third-party providers function. I assume the plugin system Gurman has described in prior reporting is separate from the Ask/Siri interface. People will be able to talk to the Apple chatbot, which has all of their information and apps from the Personal Context and App Intents, but they’ll also be able to use ChatGPT instead if they prefer. It’s not like OpenAI will be a secondary model provider for the Siri interface — those models are bespoke, custom-designed Gemini models commissioned by Apple. Clearly the Siri chatbot will be the main affair, and secondary chatbots will be able to chat with users through the same interface.

I will note that I’ll find it odd that there’s even any doubt that Apple won’t ship these features in Beta 1. The company literally settled a class action lawsuit alleging that it misled customers about Apple Intelligence features in 2024 — the same features that still haven’t launched. In the post-Something Is Rotten in the State of Cupertino era, I think the world would completely lose it if Apple dared to come up onstage and say the new features are again coming “later this year.” It’s Beta 1 or bust — I don’t reckon anyone is willing to give Apple more timeline leniency.

Aaron Tilley, reporting for The Information:

As part of an Apple agreement with Google, some user queries to a new version of Siri will run in Google Cloud on a licensed version of the search giant’s Gemini model. Apple recently approved the use of a privacy technology from Nvidia in that setting, suggesting it will use Nvidia AI chips for at least some of its computing needs in Google Cloud, according to people familiar with the matter.

Specifically, Apple will tap into Google’s fleet of Nvidia’s Blackwell B200 data center chips, said the people. Apple will enable Nvidia’s confidential compute feature that encrypts data as it’s being processed on the chips.

The Information’s report also says that it is “unclear” how Apple will use Private Cloud Compute to run the white-label Gemini models, but I think the ambiguity is all the reporting we need. I contrarily predicted in January that Apple wouldn’t abandon Private Cloud Compute because it “wouldn’t squander the last remaining competitive advantage it has in the AI space,” and I stand by it. I don’t think Apple will hand all inference off to Google. While these Nvidia processors are fast and efficient at running inference, Apple has made lofty promises about the carbon neutrality and privacy of its AI products. This was a defining feature of the announcements two years ago. To think that Apple would abandon those virtues entirely at a time when backlash to AI is so outsized is foolish.

I’m not sure how Apple will split inference between Google’s servers and its own, but I imagine that the company will handle most inference either on-device or with Private Cloud Compute, not Nvidia and Google’s privacy stack. Maybe only the most difficult prompts will be sent to Google Cloud, or perhaps there’ll be an option to enable third-party inference in exchange for less privacy assurance. Something like “Allow Third-Party AI Processing.” Using Private Cloud Compute and touting 100 percent carbon neutrality during one of the most heavily watched technology presentations of the year would be an indescribably gargantuan boon to Apple’s AI efforts. If Apple came out and said, “Your data is confidential, we don’t train on it, it doesn’t kill the planet, and it doesn’t require a data center in your backyard,” I’d imagine that’d be one of the best AI pitches out of Silicon Valley.

Apple is undoubtedly using Google Cloud for training, since it has delegated that part to Google, but more efficient inference is what people seem to care the most about. Two years later, I hope Apple has figured this out.