The decision that shaped Apple’s AI at WWDC 2026 went unannounced: there is no AI app. Apple ships no chatbot to open and enters no race for parameter counts. What it ships instead is a model spread across Siri, Camera, Spotlight and Shortcuts, plus two APIs that let any third-party app join that circuit.
It is a different bet from the rest of the industry, and it plays out across iOS 20, macOS 17, watchOS 11 and visionOS 3.
Siri Becomes Context-Aware
The headline feature is a context-aware Siri that understands relationships across apps. Ask “When is my concert and what’s the fastest route there?” and Siri surfaces the ticket confirmation from Mail, cross-references the venue in Maps, checks live traffic, and delivers a unified answer with a suggested departure time.
Siri can also recognize people, objects, and places in photos. In one demo, pointing an iPhone at a concert poster triggered Siri to identify the band, pull up their latest album in Apple Music, and offer to add the tour date to the calendar. “This is Siri as it was always meant to work,” said Craig Federighi, Apple’s senior vice president of software engineering.
Describe a Shortcut: Automations in Plain English
Shortcuts has long been a power-user tool requiring familiarity with its visual action builder. WWDC 2026 introduces Describe a Shortcut, a natural language interface that lets anyone build automations by typing or speaking what they want.
“Every morning, tell me the weather, open my garage door, and play NPR in the kitchen”: Shortcuts parses the intent using Apple’s on-device language model, maps it to available actions, and assembles the complete workflow. Power users can still tweak generated shortcuts, while the barrier to entry is gone.
Camera App Gains Visual Intelligence Mode
The iPhone Camera app now includes a dedicated Visual Intelligence mode alongside photo, video, and portrait controls. Point the camera at a real-world object and ask questions aloud. Siri identifies the subject, retrieves relevant information, and reads a summary without leaving the viewfinder. Common objects are recognized entirely on device; complex queries fall back to Apple’s Private Cloud Compute.
Spotlight AI Search on Mac
macOS 17 brings deep document understanding to Spotlight. The system-wide search tool now analyzes the contents of PDFs, Word files, and presentations, answering natural-language queries like “Find the slide about Q3 revenue” or “Show the email where Sarah discussed the budget revision.” The results match the contents of files instead of their names. Local indexing uses Apple’s on-device model, keeping data on the machine; web queries consult Apple’s cloud models with privacy guarantees.
Apple Foundation Models: Built from the Ground Up
All these features are powered by Apple’s family of foundation models, detailed in a research paper published alongside the keynote. The on-device model runs approximately 3 billion parameters, sized for A18 and M5-series chips, and was trained using AXLearn, Apple’s open-source framework built on JAX and XLA.
The model uses a LoRA-style adapter architecture that swaps task-specific adapters (one for Siri, another for Spotlight, another for Shortcuts) without loading separate models, keeping memory footprint low. When a request exceeds the on-device model’s capacity, Apple’s Private Cloud Compute (PCC) infrastructure, a dedicated fleet of Apple Silicon servers, handles it in a privacy-preserving, cryptographically attestable environment. PCC, first announced at WWDC 2024, is now production-ready and handling millions of daily requests.
Google Gemini Integration
In a pragmatic move, Apple announced that Siri can route certain requests to Google Gemini when neither Apple’s on-device nor cloud model delivers the best answer. Users must explicitly opt into third-party model access, and Apple’s privacy architecture ensures Gemini processes queries without linking to any Apple ID.
A “model router” within Siri selects the optimal engine per query. Apple’s models handle the vast majority of requests, while Gemini is invoked only for especially complex or knowledge-intensive queries where Google’s larger parameter count provides a measurable quality advantage, and third-party access can be turned off entirely in Settings.
Developer Implications: App Intents API and View Annotations
For the developer audience at WWDC, Apple shipped two key APIs extending these AI capabilities to third-party apps.
The expanded App Intents API lets apps expose functionality to Siri and Describe a Shortcut with typed parameters, natural language examples, and dynamic responses. Apple’s foundation model uses these intent definitions to route requests to the correct app, and can chain multiple apps in a single Siri utterance.
The new View Annotations framework lets developers mark up SwiftUI views with semantic metadata: “this is a contact’s phone number” or “this text contains a flight departure time.” Siri and Spotlight index these annotations system-wide, enabling cross-app search and direct actions. “If your app shows a reservation, a tracking number, or an event, annotate the view and Siri will find it,” Apple’s developer tools team explained. Both APIs are available in beta today, with production SDKs shipping this fall alongside iOS 20 and macOS 17.
The Big Picture
Apple’s WWDC 2026 AI strategy leaves three things out: a standalone chatbot, a race to the largest parameter count, and a single AI app. Intelligence is embedded throughout the OS, in Siri, Camera, Shortcuts and Spotlight, with privacy as the differentiator. The combination of on-device models, Private Cloud Compute, and selective third-party integration (Google Gemini) gives Apple flexibility without compromising its stated values. For developers, the App Intents API and View Annotations framework offer a clear path to make apps discoverable and actionable through system-level AI, with no separate AI SDK required. Whether this marks the moment Apple leapfrogs its AI competitors remains to be seen, but one thing is certain: after WWDC 2026, Siri is no longer an afterthought.