By Jason Snell

June 10, 2026 12:40 PM PT

WWDC 2026: Emptying the notebook about AI, bug fixes, and more

I’m home after two and a half days down in the South Bay for WWDC. This year I tried to take notes at the keynote and in briefings using a traditional reporter’s notebook and a pen, which means that I’m literally emptying my notebook when I give you these first scattered impressions of what I saw.

The AI stuff

What struck me most about Apple’s AI announcements is how little the company’s stance on AI has changed since its fateful WWDC 2024 announcements. It’s still classic Apple: AI is not the end but a means to an end, with the goal of building “helpful products for people.”

So this release is not about Apple questioning its priors. Instead, it’s about getting back on track, back in the game. The goal here isn’t for Apple to blow away the competition, but to be relevant and helpful and create a foundation on which to build.

Also, having a version of Siri that actually works would be a pretty big win.

Apple’s huge advantage here is that it’s the platform owner, so it can build tools that search through all the data on your personal devices without requiring that you expose that private data to some company’s systems. It’s using that revamped Spotlight to search through your own data, then handing snippets off to Private Cloud Compute for processing. I think it’s all very promising.

In terms of where it’ll go next, look to the Passwords feature that agentically changes your passwords for you in the background. Look at the tools that let you vibe code Safari extensions and create Shortcuts. I do not doubt that Apple would love to do more of this sort of stuff, which is much closer to the present-day of AI enthusiasm, but it really does need to walk before it can run.

More broadly, I’d like to see Apple ship these features this fall and then maybe introduce some new AI features early next year. The pace of AI is so much faster than Apple’s annual software cycle can accommodate. It needs to get used to phasing in new AI features across the entire cycle, so it can make quick adjustments as new trends in AI functionality emerge.

I want to note that this year Apple has redefined what Private Cloud Compute, a concept it introduced in 2024, means. Before, it meant Apple servers running on Apple hardware in Apple data centers. Now it means something a bit broader, since it can also include Apple-controlled servers running on non-Apple hardware in Google data centers. Apple took great pains this week to explain that Apple controls those servers and they’re built to the same privacy specifications as the other servers in the PCC cloud—in other words, they’re not generic Google servers that could compromise your private data—but it’s also a sign that Apple needed more cloud AI power than it was capable of providing on its own. Hence the redefinition.

And one final AI note: The segmentation of AI models has commenced. Apple now has two different on-device AI models, one of which has much higher hardware requirements. Right now, this higher-powered model is primarily used for improved dictation and speech synthesis, but undoubtedly over time, it’ll be used for other things. I do wonder if, in the long run, older Apple devices will just have to turn more to Private Cloud Compute to perform beefier tasks, or if they’ll be entirely barred from new features? But we’ve already seen that being a device “capable of running Apple Intelligence” is no longer sufficient for some features.

On the server side, there are also multiple models. More basic jobs are handed to a smaller model running on Apple’s servers. Heavier tasks are instead handed to the bigger models running on Apple-controlled servers in Google’s data centers. This is all transparent to the user, which is as it should be, but it’s interesting to watch Apple’s AI back-end increase in complexity.

Snow Leopard revisited

The moment the keynote used the phrases “sweating the details” and “attention to detail,” it was clear that beyond AI and Siri features, this year is about small fixes and improvements. In more private settings, Apple folks specifically referenced Snow Leopard and iOS 12, two updates that saw Apple take a pause from huge feature roll-outs and prioritize speed and bug fixes a bit more.

But to be clear, referencing Snow Leopard does not mean “no new features.” Like Apple’s 27 releases, Snow Leopard was full of dozens, if not hundreds, of new features—they were just scattered small improvements throughout the system. Updates like this are a challenge to communicate because there are no big features to grab on to.

This is the conundrum of operating-system releases. People say they want bug fixes and small quality-of-life improvements, but a roll-out without tentpole features feels kind of bland. In any case, Apple’s all-out mission to finally fix Siri and get AI integrated in their products the way it said that it would two years ago has allowed the rest of the people developing software at the company to check a bunch of items off their longstanding to-do lists, and I’m here for it.

At the top of my “why did it take them this long?” list of improvements is the change, mentioned in the keynote, that will allow iPhones to better handle that moment when they leave an area with Wi-Fi and have to switch to cellular. I always think of this as the “driveway problem,” but whatever you call it, too often I’m sitting in my driveway looking up directions in Apple Maps only to have all my searches fail because I’m apparently too far from my home Wi-Fi, but my iPhone hasn’t given up hope that it’ll come back soon.

My understanding is that in iOS 27, the iPhone will rely not just on measurements of signal strength (which is the primary method of choosing the wireless network today), but will also use throughput, latency, and signs of network congestion as signifiers. And it’s designed to do so quickly, so you don’t spend as much time frustrated because your iPhone feels it hasn’t sufficiently mourned the loss of its Wi-Fi signal.

Apple fixes stuff it needs to use

A lot of frustrating bugs sit, untouched, for years. Apple has its priorities, and shiny new features get the love while rickety old stuff never rises to the level of being important enough to fix.

Until, that is, Apple needs to have that feature work right in order to serve one of its priorities for the latest OS release. At that point, you’ll find that old, broken features suddenly get the attention and fixes they’ve needed for years. That’s why some of the seemingly random big fixes and improvements scattered across the 27 releases aren’t actually random! They’re side effects of Apple’s larger feature pushes.

For example, imagine that you’re building a new Siri AI system that needs to lean on searching through a user’s local files in order to apply an important level of personal context. Perhaps when you’re building that system, you realize that you can’t actually rely on Spotlight to supply all of that context because it’s not nearly as stable or efficient as you need it to be.

If such a thing were to happen, well, perhaps Apple would find the time to rebuild all of Spotlight search to make it work faster and more reliably. Perhaps searching in Mail would float more relevant results up to the top. Perhaps Messages search would become less frustrating. And perhaps users who need to search for things will benefit, even if they’re not heavy users of Siri AI itself.

What I’m saying is, Spotlight’s going to be better in the 27 releases.

Long-suffering Shortcuts users will notice similar things happening there. It’s been incredibly frustrating to develop Shortcuts due to the lack of support for a proper If-Else statement, a cornerstone of programming. Scheduling Shortcuts has also been a pain, because you’ve got to tie a shortcut to a separate Automation step.

Well, guess what? Apple is introducing a cool new feature that lets you build Shortcuts entirely out of text prompts, using an AI model. It works pretty well, at least for basic tasks, and I’ll have a lot more to say about it this summer. But I have zero doubt that the people building that feature looked at Shortcuts and said, essentially, “What do you mean it doesn’t have If-Else or integrated scheduling? We need those things!” And so they’re now going to be there, for all Shortcuts users to take advantage of.

visionOS: Not dead

People are quick to bury the Vision Pro, which is and has always been a speculative and impractical device that’s more about the future than the present. It’ll be years before Apple is able to construct anything like the Vision Pro at a price and with a feature set that could possibly make it a mainstream product. In the meantime, it’s an experiment and exploration, and I’m okay with it. For all that to be true, though, visionOS needs to keep advancing. And it looks like it is.

This year, Apple’s adding the ability to convert panoramas into spatial scenes, which is just a wild idea. I’m dubious that my Sligachan panorama from the Isle of Skye is going to replace Bora Bora or Joshua Tree, but I love that Apple is still tinkering—and panoramas and spatial scenes are some of the best features in the Vision Pro. (That’s also a good sign, because it suggests Apple is learning what works well on visionOS and is leaning into those features.)

Also, if you believe the stories, Apple exec Mike Rockwell—who is the guy who was charged with shipping the Vision Pro—had originally planned for visionOS to be much more driven by Siri, only to be repeatedly let down by the Siri team. In visionOS 27, Siri AI seems to be pervasive. It feels like Rockwell, who is now in charge of Siri, is having his revenge—and fulfilling one of his dreams for how visionOS should work.

Of course, visionOS is also a playground for future features of other Apple devices. As Dan has pointed out, some of the visual-intelligence features of visionOS 27 sure feel like they might be applicable to other future wearable devices that Apple might be working on. Again, visionOS being a platform for experimentation is a good thing.

Photos improves, but it’s complicated

As someone who writes a book about the Photos app, I’m extremely invested in the changes Apple makes to that app from year to year. This year, it’s addressing one of its biggest limitations and adding a load of AI features that I’m ambivalent about.

First, the good news: Changes to Shared Albums! This feature has been compromised since it was launched, since it didn’t offer the ability to share full-resolution images from your library. Over the years, Apple added other methods of sharing groups of photos via iCloud, and those could include full-resolution images, but this one prominent feature felt stuck in the past.

Now it’s getting a proper upgrade, with support for full-resolution images and allowing for full collaboration with people on other platforms so that everyone can contribute to a shared photo album. I’m relieved that I will soon have to stop explaining the differences among the various ways to share items in Photos and warning people away from Shared Albums.

As for the three AI-powered features in Photos, they’re a mixed bag. I have high hopes for a much improved Clean Up, which was already okay but could be a lot better. The new version appears to be much more adept at artfully clipping unwanted items out of an image and filling those areas with in-context imagery. This is where generative AI is really required, because if the fill-in algorithm isn’t smart, the results will look fake. So far as I know, Clean Up occurs on your device, using on-device models.

The other two features, Extend and Spatial Reframe, require the use of an advanced diffusion model that’s only available via Private Cloud Compute, and as a result, they take time to execute, since Photos will need to upload your photo, wait for a result, and then download the result.

Extend feels like a good feature, since there are plenty of scenarios where your image needs just a little more headroom or width. It’s also going to be great for straightening images, since Extend can fill in the slivers of unknown image that are exposed when you rotate, which otherwise require that your image be cropped as you rotate.

However, every pixel you expand the selection increases the jeopardy that what’s going to be generated is weird or fake. Everyone’s mileage may vary, but I found that I was much more comfortable expanding a photo a little bit to gain some headroom than doing it a lot, forcing the AI system to invent more objects or scenes. Judicious use would be my recommendation.

Then there’s Spatial Reframe, which brings together a load of existing Apple technologies, including the spatial scanning algorithm it used to create spatial photos on Vision Pro. This feature works by scanning your photo locally using that algorithm, inferring a depth map that is then used to build a 3-D version of the image that you can pivot a bit, up and down and left and right. This is the effect that allows you, on the Vision Pro, to feel like you can move your head and see parallaxes shifting, even though, if you look closely, the exposed content behind a subject is just a simple generative fill. It all happens so quickly, and in service of a live 3-D effect, that it’s often not noticeable, and even when it is, it’s not that big a deal.

The bar is a lot higher for a fixed, 2-D photo at full resolution. So after you use Spatial Reframe to slightly move the perspective of a shot, all the data is sent up to Private Cloud Compute, where a new version of your shot is rendered—including much more advanced generation of all of those pixels that are revealed by parallax or at the edges of the frame.

The problem is that these results feel pretty generative, through and through. I saw some samples of people’s faces that, after being Spatially Reframed, didn’t really look like their faces anymore. Unlike Extend, Spatial Reframe changes the entire perspective of the picture, which requires everything that’s visible to be re-rendered at full quality. The result is an image that, at least based on my initial reactions, felt surprisingly artificial. I’ve got to use this feature a lot more over the summer, but my initial reaction is skepticism.

If you appreciate articles like this one, support us by becoming a Six Colors subscriber. Subscribers get access to an exclusive podcast, members-only stories, and a special community.