The undisputed star of Apple’s WWDC26 keynote was “Siri AI.” Apple did away with the detailed introductions to individual operating systems such as iOS and macOS that had long dominated this developer conference, focusing instead on the revamped Apple Intelligence and the radically different role it would give Siri. It really is coming out this time, right?
At first glance, the Apple Intelligence announcement might leave you thinking, “Wait, don’t we already know all these features?” or “Can’t other smartphones already do this?” So some of you may have found it a little underwhelming. At the same time, people deeply invested in the Apple ecosystem probably welcomed the fact that the role they had long hoped AI would play on the iPhone and Mac was finally becoming real.
Let’s take a closer look at what WWDC26 revealed, why Apple’s AI has taken so long, and what Apple is trying to do with it.
Apple Intelligence: What happened to the promise of 2024?
Time really does fly. Apple first unveiled Apple Intelligence to the world at WWDC in 2024, which means two years have already passed. Looking back, AI at the time carried enormous expectations: it seemed ready to move beyond mere potential and do something truly extraordinary.
Yet Apple itself had been rather quiet right up to that point. Sharp criticism poured in from the market: “Why isn’t Apple putting a massive large language model, or LLM, front and center like OpenAI, Google, and Meta?” Even for a company that has always kept so much under wraps, Apple seemed unusually quiet.
In truth, Apple had probably been waiting for this AI-centered shift as eagerly as anyone. It was Apple, after all, that introduced the Neural Engine and got an early start on machine-learning developer tools to prepare for the AI era. But ChatGPT brought an enormous, sudden wave from a different direction. AI was heading somewhere so different from what Apple had long envisioned that the company seems to have faced some difficult questions.
This went beyond a technical problem. It clashed head-on with Apple’s philosophy of putting personal information and privacy first.
In the end, Apple chose not to build an AI that would embrace all the knowledge in the world. It chose a “personal assistant of my own,” built around its devices and woven deeply into everyday life. That decision aligned perfectly with the unease ordinary people feel about entrusting all their personal information to AI.
To achieve this, Apple decided to do most AI processing inside the device itself, or “on-device.” The logic was that, paradoxically, only by keeping personal information safe could it gain access to a user’s most intimate personal information. That decision made sense to me then, and it still does. Ordinary access to personal information is worrying enough. Combine it with AI, and who knows what the consequences might be?
Apple’s goal at the time was clear: learn from and remember all the data on a device, understand the broader “context” connecting all that personal information, and use Siri to engage more deeply with my everyday life.
The bitter cost of privacy and local AI
But the price of that decision was far from small. Devices first needed enough performance to run this demanding AI, and a shortage of memory immediately got in the way. Apple ended up having to launch this enormous AI platform within an “8GB of memory” constraint, creating the rather absurd situation in which, at the time, it ran only on the iPhone 15 Pro lineup. The iPhone 15 tells us that even Apple had not anticipated AI developing to this extent.
Developing the AI models themselves was also a thorny path. Machine-learning-based AI matures over time through training on and validating vast amounts of data, and Apple did not have much time. On top of that, it had set a firm principle that user data would never be used for training. Just assembling clean, legally usable training datasets must have been an enormous challenge. Even for Apple, putting it all together in a few months would have been close to impossible.
What’s more, this was a period of upheaval in AI: OpenAI was growing explosively, Google was fighting back, and new heavyweights such as Anthropic were emerging. The world’s attention, talent, and infrastructure were all flowing toward the giant cloud AI systems of global Big Tech. Even looking back now, Apple’s strategy of moving inside mobile devices with severely limited performance and data makes me wonder whether it was really feasible.
Ultimately, the first generation of Apple Intelligence failed to deliver the language-model performance people had expected, and Siri’s major evolution was postponed. I thought something big would happen at last year’s WWDC25, but Apple simply kept repeating, “Please wait a little longer.” This was not merely a question of strategic direction. It was an immediate, real-world crisis in which Apple risked losing control over the mobile experience.
The beginning of a shift: Joining forces with Google
In the meantime, all sorts of stories circulated, and none of them were good. Then, early this year, Apple finally joined forces with Google and began building a new foundation model for Apple Intelligence. In some ways, it was a very natural next step. From the outset, Apple had designed its system to leave the world’s vast store of information to outside services such as ChatGPT, and it was ready to accommodate any AI platform flexibly. That was not Apple’s core competitive territory. But improving “Siri’s core brain,” the part that accurately understands the context of my personal information, had become an urgent priority.
And one of the companies furthest ahead in this field was Google. What has Google’s AI always done best? It reads the context linking users’ personal information and activity logs to anticipate what someone likes right now, what they want to buy, and where they want to travel. A kind of mind-reading, if you will. This is a core technology directly tied to Google’s advertising revenue. More recently, Google’s Gemini has also been considered one of the AI models best at understanding what people mean and making sense of the world. Google, in other words, had everything Apple was desperately looking for.
Above all, the two companies have worked together for a long time and know each other’s technical direction well. Compared with OpenAI or Anthropic, which Apple knew relatively less well, it probably saw Google’s approaches to AI, including data transparency, as more openly documented. Fundamentally, the two companies use different methods to handle AI and data from a privacy perspective, but their principles and direction are similar.
The foundation model in this new, second generation of Apple Intelligence was developed by extensively incorporating Google’s Gemini. That does not mean Apple simply took Gemini and used it as-is. Apple remains firmly committed to its original promise and guiding principle: protecting personal information. It has built its own “hybrid” model, processing what it can on the device and, when more performance is needed, using a tightly secured cloud only to provide additional computing. Naturally, Siri AI’s new role and character are built into this as well.
Senior vice president Craig Federighi made a rather telling remark in this keynote, too: “AI is a powerful technology that does meaningful things, but I’m concerned that the world is charging ahead without looking around.” I strongly agree. AI is so convenient that I use it more and more, and before I know it, these services seem to know almost everything about me. You ask a simple question, and it connects your personal information with your recent interests to suggest, “How about trying this?” It is helpful, but sometimes it sends a little chill down your spine.
Where Siri AI is headed: Evolving into AI that takes action
This Apple keynote focused entirely on expanding the evolved Apple Intelligence and showing what Siri AI can actually do. Apple Intelligence now understands language far more smoothly and actively incorporates the latest information from the wider world to produce intelligent answers. It has properly absorbed the key elements of today’s popular LLMs.
Image processing has become dramatically more powerful, and, above all, Siri’s speech is much more natural and emotionally expressive. Its intonation and emotional tone have become so natural that they bring Google’s Gemini Live to mind.
And, just as Apple promised two years ago, Siri understands my personal information as a whole and pieces together answers from it. What felt distinctly different this time was how the data is organized. Apple did not spend much time on the operating systems’ visual changes in this keynote. In fact, even if you install the current OS betas, there are no particularly major visual changes.
Instead, Apple appears to have overhauled the entire underlying structure for processing data. It has streamlined the internal scheduler to make programs run faster, and loading and transferring data have become dramatically quicker. Searches are much more accurate, too. This is clear evidence that, behind the scenes, Apple Intelligence is learning from and organizing data far more actively and systematically than before.
Now, looking at the app-integration examples demonstrated at this WWDC, one very interesting common thread emerges. It may be Apple’s most important message from the entire event: “action.” Shall we revisit the keynote demos?
https://youtu.be/hF8swzNR1-o?t=2161
While looking at a photo, you ask Siri, “Where exactly is this?” Siri tells you the location. Then you ask, “Jeff said he moved recently. Where did he say he lives?” Because Siri knows all the messages and emails you have exchanged with Jeff, it finds the new address in an instant. You follow up with, “Then give me directions to Jeff’s house, and from there to the place in that photo.” Siri immediately opens Apple Maps and shows the navigation route.
Next, you ask about the World Cup schedule. Rather than merely responding in text, Siri presents each country’s match schedule in neat visual cards, much like recent knowledge-graph technology. You pick a particular match and ask about signature dishes from the two countries, and Siri explains them readily. Finally, you say, “Send an invitation to the group chat with a menu that includes these dishes and the World Cup match schedule.” Siri opens a message window with all that information and asks you to confirm. All the user has to do is press Send.
Even the example of calling an airline is interesting. In the brief moment while the phone is ringing, Siri recognizes that the number belongs to an airline and works out why the user is calling. Usually, it would be to confirm or change a booking, right? Siri immediately searches my email and proactively displays key details such as the flight schedule, flight number, and booking reference in a pop-up on the screen. There is no need to scramble for information once an agent answers.
Ultimately, Siri AI’s destination is to “take action” on the user’s behalf. Take this a step further, and we call it “agent AI.” It is part of the same trend as the open-source OpenClaw, which recently triggered widespread Mac mini shortages, or Google’s Gemini Spark, which actively gets work done as if a person were using the keyboard and mouse.
In fact, this agent AI capability is exactly what I am most looking forward to from Siri. But today, agent AI inspires enormous concern as well as excitement. Put simply, there is the fear that “you never know what this thing might get up to behind the screen.” Give it almost every permission in the system, and one wrong move could mean paying for the wrong item, sending my sensitive personal information to the wrong email address, or even wiping out my stock trading account.
That is why what we ultimately need is “reliability and complete trust.” This is precisely why Apple has continued to insist on privacy and delayed adopting AI, even in the face of criticism. Only with complete trust as a foundation will users eventually feel comfortable giving AI substantial authority to act.
Ultimately, Apple’s AI advantage comes down to trust
Of course, even this feels a little premature right now. At the recent Google I/O, Google put agent AI at the center of its efforts to build powerful capabilities for handling personal information. Apple, by contrast, remains cautious, and its current models still have a long way to go before earning users’ complete trust. AI needs more than technical maturity. The emotional sense of security users feel matters, too. I think it is fair to say that Google and Apple are each approaching this in their own way.
But this time, Apple sent a clear message about where it intends to go. After understanding context through personal information and reading the user’s intent, the next step is ultimately “action.” This is the moment when the trust Apple has worked so hard to build connects directly to “AI that takes action.”
It has certainly taken a long time. And unfortunately, it looks as though we will have to wait a little longer. Initially, the new Apple Intelligence and Siri AI will be available only in English in the United States. This restriction is expected to remain in place not only for the developer beta but also at the official launch this fall. Apple has, of course, officially promised Korean support. Fortunately for us, various regulatory and political issues mean Europe and China will not be able to use it immediately, so our turn may come a little sooner. Even so, it will probably be around the end of this year before we meet a truly capable Siri that speaks our language, Korean, fluently.
There is plenty to feel disappointed about with this round of Apple Intelligence, and plenty left to wish for, but I am certainly looking forward to it. Many of you probably already use powerful AI tools such as Gemini, ChatGPT, and Claude in everyday life. But using them as third-party apps is an entirely different experience from having them work seamlessly and reassuringly deep within the operating system of the smartphone I use every day. Everyday AI that lets us nod and say, “Good, you really do understand,” rather than anxiously wonder, “Wait, how did it know that?” will give us an entirely new experience.
We could argue about whose underlying AI technology is better or worse. But the point is that all we need to do is make good use of these excellent services together inside our smartphones, tablets, and computers. Apple has never been a company that invents every underlying technology itself. It takes good technologies from around the world and turns them into easy, convenient experiences in its own way.
AI is no different. Google’s excellent AI model is simply another of the “best ingredients” that go into an Apple device, just like LG’s displays or Samsung’s memory. Now, I think we can enjoy watching and looking forward to the new user experiences Apple will create with these outstanding ingredients.
At last, the direction and the right ingredients are in place, and devices with the powerful processors to support them are making their way into our hands. Now it is time for Apple to show us AI in a truly Apple way.
