Apple finally gives Siri the AI rebuild it should’ve had two years ago

Apple used WWDC 2026 to show the Siri overhaul it promised two years ago, then failed to ship. The new AI-powered Siri arrives in beta later this year, with Apple pitching it as a broader assistant across chat, writing, on-device context, and system control.

That’s a big change for a product that has spent years handling timers, weather, messages, music, and smart home commands. The new Siri is supposed to answer open-ended questions, draft text, reason over personal data, understand what’s on screen, and handle a back-and-forth conversation.

Apple is also giving Siri its own app, while keeping it built into iPhone, macOS, and watchOS. On modern iPhones, Siri moves into the Dynamic Island with a new animation. Users can swipe down from the Dynamic Island to search or type to Siri, press the side button, or still say “Hey Siri.”

The comparison with ChatGPT, Claude, and Gemini is obvious. Apple’s edge is OS access. If the permissions model works, Siri can reach private context scattered across messages, calendar events, email, contacts, and whatever is visible on screen.

Developers and technical teams should pay attention to that part.

Siri gets serious about context

The headline feature is context. Apple says the new Siri can use current world knowledge, but also pull from information on the user’s device. That includes what’s displayed on screen, previous emails, calendar details, contacts, text messages, and web information.

In practice, a user could ask Siri to draft a message that references a meeting from their calendar, a point from an email thread, and a document currently open on screen. Or they could ask for a travel plan that considers flight details from Mail, dinner plans from Messages, and a contact’s address.

That’s where phone-based AI assistants start to get useful. General-purpose chatbots are good at synthesis, but they usually need users to paste in the context. Siri can skip that step if it can safely access the right data at the right time.

Apple showed text cards for answers, including web results and information pulled from messages. That interface choice matters. Voice-only responses are bad for dense answers. Cards give Apple room to show citations, entities, extracted snippets, and actions without forcing everything through speech.

The unresolved technical question is where the work happens. How much runs locally? How much goes to Apple’s servers? How often are third-party model providers involved? Apple’s WWDC framing focuses on personal context, but that’s also the hardest data to process safely. A context-aware assistant is only as trustworthy as its data boundary, prompt isolation, and permission design.

For AI engineers, that’s the interesting work. Generating a friendly paragraph is the easy part. The hard part is deciding which personal data is relevant, retrieving it accurately, avoiding accidental disclosure, and grounding the response in a way the user can inspect.

The Siri app changes the product

Apple is launching a dedicated Siri app alongside the assistant. That’s notable because Siri has usually been an ambient feature, not somewhere users go intentionally.

A standalone app gives Apple a place for longer conversations, brainstorming, writing feedback, and multi-turn planning. It also gives users a chat-style surface that competes more directly with ChatGPT and Gemini. Users can ask for an in-depth plan, go back and forth, brainstorm ideas, or get comments on a document.

That’s a practical change. Many AI tasks don’t fit the old Siri model. “Set a timer for 12 minutes” works as a one-shot voice command. “Help me prepare a response to this client based on the last three emails and my calendar availability” needs a workspace, visible context, editing, and sometimes correction.

Apple is keeping Siri available from system entry points, while using the app for longer AI interactions.

The risk is fragmentation. If users have a Siri app, Dynamic Island Siri, Spotlight Siri, dictation, “Write with Siri,” and app-specific AI controls, Apple has to make the mental model clear. Otherwise, people won’t know which surface to use, or why one produces better results than another.

“Write with Siri” is the obvious productivity play

The new “Write with Siri” feature brings AI-assisted drafting to Mail and Messages. Apple says Siri can match how a user typically communicates with a specific colleague or friend. If you usually send your manager short, direct bullet points, Siri should draft in that style.

That sounds useful. It’s also easy to oversell.

Style matching depends on access to previous messages, accurate classification of tone and relationship, and a model that doesn’t flatten everyone into the same bland corporate register. Developers who’ve built writing assistants know this failure mode well. The model often captures surface patterns, such as bullets or length, while missing the social context that makes a message appropriate.

Apple does have an advantage here. Mail and Messages contain the signal. If Siri can summarize communication patterns locally or with strong privacy controls, it could produce better drafts than a generic chatbot that has never seen the user’s actual writing.

The trade-off is consent. Users need to understand when Siri is analyzing past conversations to imitate tone. That kind of personalization can feel helpful in one moment and invasive in the next. Apple will need clear controls, especially in professional settings where messages may include sensitive company data.

Enterprise teams have policy work to do as well. If Siri can synthesize from Mail, calendar, contacts, and messages, device management policies need to catch up. Companies will want to know whether generated drafts can include confidential information, whether prompts leave the device, and how auditability works.

Spotlight starts answering questions

On macOS, Apple is integrating Siri into Spotlight. Users can tap into Siri from anywhere, and Spotlight can now find answers to questions, not just files, apps, and settings.

That’s a logical move. Spotlight already sits at the center of Mac navigation. Turning it into a question-answering layer could make macOS feel less like a file system plus app launcher and closer to an indexed personal workspace.

The implementation details matter. A useful AI Spotlight needs to rank local files, app data, emails, calendar records, and web results without burying the user in plausible nonsense. If it returns a synthesized answer, users need to see where that answer came from. Retrieval-augmented generation works best when the retrieval layer is transparent enough to debug.

Senior developers should care because this points to a broader platform shift. Operating systems are starting to expose semantic search and action-taking AI as native capabilities. If Apple eventually opens richer APIs around Siri intent handling, app entities, and screen context, developers may need to structure app data so the system assistant can understand it.

Apple hasn’t detailed those developer hooks in this announcement. The direction is still clear: apps that keep meaningful state trapped in custom UI without accessible metadata may become less useful in an assistant-driven OS.

Voice helps, but latency will matter more

Apple says Siri has a new voice experience, with customized pace and expressivity. System-wide dictation is also getting an accuracy improvement, including better spelling, punctuation, and capitalization.

Dictation improvements are easy to overlook, but they matter for daily use. If Apple can reduce correction time, voice input becomes a better interface for longer prompts and messages. That matters on iPhone and Apple Watch, where typing is often the bottleneck.

On watchOS, users will be able to ask questions and take action directly from the watch. That’s a natural fit for quick tasks, but it puts pressure on response latency. A watch interaction has almost no patience budget. If Siri needs several seconds to retrieve context, call a model, and generate an answer, users will give up.

Performance will decide a lot. AI assistants feel useful when they respond quickly and correctly. They feel broken when they pause, misunderstand, or give a confident answer that misses obvious context. Apple’s hardware and OS integration help, but multi-source retrieval and generation are still expensive operations, especially when privacy constraints limit server-side processing.

Apple has integration. Siri still has to earn trust.

Apple is late to this version of the assistant market. ChatGPT, Claude, and Gemini have trained users to expect high-quality conversational answers, code help, document summarization, and planning. Siri has trained users to expect disappointment.

That history matters. Apple can ship a technically improved Siri and still struggle to change behavior if early beta interactions are flaky. Users don’t give assistants many second chances.

The new Siri has three clear advantages:

• It can access system-level context that standalone chatbots usually can’t.
• It can act inside core apps like Mail, Messages, Spotlight, and watchOS.
• It can use Apple’s privacy positioning as a differentiator, assuming the architecture backs it up.

The weaknesses are just as clear. Apple hasn’t proven that Siri can handle complex multi-step instructions reliably. Personal context retrieval can fail in subtle ways. Writing assistance can create messages that sound polished but socially wrong. Screen awareness creates new privacy edge cases. Beta availability later this year also means developers and enterprises won’t know the real behavior until they can test it on actual devices.

Hallucination remains a problem. Grounding in web and device data helps, but it doesn’t remove the risk. If Siri summarizes a message thread incorrectly or invents a detail from a calendar event, the error may be harder to catch because it appears inside a trusted system interface. Apple will need strong source display, confirmation flows for sensitive actions, and conservative defaults when Siri acts on behalf of the user.

For developers, the thing to watch is how Apple exposes app context to the assistant. If Siri becomes a primary interface for finding, summarizing, and acting on app data, app architecture may need to account for machine-readable state, permissions, intents, and semantic indexing. Productivity apps will likely feel this first, followed by messaging, project management, CRM, notes, health, and developer tools.

Apple has put Siri on the same playing field as modern AI assistants. The beta label is doing a lot of work, though. The demo version sounds like the assistant Apple should’ve shipped years ago. The shipping version has to prove it can reason over messy personal data without being slow, creepy, or wrong in expensive ways.