WWDC 2026: Apple Intelligence Goes Gemini, and the Developer Stack Gets an AI Overhaul

Apple opened WWDC 2026 today with a concession few expected — and a developer story bigger than the keynote let on. Apple Intelligence now runs on Google's Gemini models, Siri has been rebuilt from scratch around it, and the frameworks you use to ship AI features — Foundation Models, a new Core AI framework, and an agentic Xcode — got their most substantial overhaul since Apple Intelligence first debuted.

Developer betas for iOS 27, iPadOS 27, macOS 27 ("Golden Gate"), and visionOS 27 are available today. Here's what's new, and what it actually means if you build for Apple platforms.

Apple Intelligence is now powered by Gemini

The headline for developers isn't a feature — it's a supplier. "This year we embarked on a deep collaboration with Google, leveraging the technologies behind their Gemini family models," Craig Federighi told the audience. Apple Intelligence still works the way Apple has described before — on-device models for lighter work, Private Cloud Compute for heavier requests, a system-wide orchestrator routing between them, and a promise that neither Apple nor third parties can read your data — but the intelligence underneath is now Gemini-derived.

Read between the lines and this is Apple competing where it's strongest — integration, privacy, and distribution — rather than on raw model research. For developers, it means the models your app reaches through Apple's frameworks are meaningfully more capable than the homegrown ones they replace: multimodal across text, image understanding and generation, and speech, without you ever choosing or managing a provider.

Foundation Models: multimodal, extensible, and finally server-side

The Foundation Models framework — Apple's API for tapping the system model directly — picked up the three things developers have been asking for since it shipped:

• Image input. You can now pass images alongside text into the model, unlocking on-device vision use cases (describe, classify, extract) without bundling your own model.
• Custom skills. A new extension mechanism lets you teach the model app-specific capabilities.
• Server-side execution. Previously on-device only, Foundation Models can now run requests server-side — the escape hatch for workloads too heavy or too large for the local model.

Apple also introduced a new Core AI framework, though it deferred the specifics to the sessions, plus an Image Playground API that exposes the system's image generation and editing to third-party apps. Pricing for server-side model calls wasn't announced on stage — watch the developer sessions closely, because that number decides whether the server-side path is actually viable for your app.

Xcode goes agentic

Federighi called Xcode "the best place" to build apps using agentic coding, and Apple clearly wants its own IDE to be the answer to Cursor and Copilot rather than ceding that workflow to outside tools. The built-in coding assistant now:

• runs agentic coding workflows instead of just autocompleting;
• handles app localization end to end;
• can interact with simulated devices to exercise what it builds;
• is extensible through custom skills.

App previews also got more interactive — easier to resize and manipulate while you iterate. If you've moved your AI-assisted coding outside Xcode, this is Apple's pitch to bring it back home.

Siri, App Intents, and a new integration surface

Siri is all-new: conversational and multi-turn, with access to both world knowledge and your personal content, and able to write messages and email and drive apps on your behalf. It gets a dedicated Siri app with conversation history synced over iCloud, surfaces in the Dynamic Island, answers right-click context queries from Spotlight on macOS, and on Vision Pro responds to a look-and-speak gesture with no "Hey Siri" needed. A new camera mode recognizes items and can even split a bill.

For developers, the integration point is App Intents, which Apple expanded so third-party apps can expose actions to the new Siri — Apple demoed the Line app letting users ask Siri to act inside it. If your app has discrete user actions, App Intents is now the highest-leverage surface to invest in: it's how your app becomes reachable from Siri, Spotlight, and Shortcuts — and Shortcuts itself now accepts plain-language descriptions instead of step-by-step building.

The platforms: iOS 27, "Golden Gate," and a more transparent Liquid Glass

The OS lineup all moves to version 27 — iOS 27, iPadOS 27, macOS 27 (named "Golden Gate"), watchOS 27, and visionOS. Notably, iOS 27 supports iPhone 11 and newer, the widest device range of any iOS release, and Apple extended its CPU scheduler optimizations back to those older phones.

Liquid Glass, last year's design language, got refinements worth knowing if you maintain a UI: a transparency slider (opaque to fully clear), sidebars that expand to the window edge with continuous refraction, colored sidebar icons, and the material now woven into app icon artwork.

Apple leaned hard on performance numbers:

Availability and the fine print

• Developer betas (iOS 27, iPadOS 27, macOS 27, visionOS 27) are out today, June 8, 2026.
• Public betas arrive later this summer; general availability is this fall.
• The new Siri is free with the OS, with daily usage limits and more headroom for iCloud+ subscribers; the full release lands later in 2026 still wearing a "beta" label.
• Regional gaps matter: the new Siri and Apple Intelligence features are unavailable in the EU at launch (Apple cites the Digital Markets Act) and in China pending regulatory review. If you ship in those markets, plan for parity gaps.

Apple Intelligence's hardware requirements remain steep — broadly iPhone 15 Pro plus the entire 16 and 17 lines, iPads with A17 Pro or M1 and newer, any M1+ Mac, and Vision Pro — so the AI surfaces you build on still won't reach your whole install base.

What it means for developers

Strip away the consumer features and WWDC 2026 is a platform-strategy statement. Apple handed the model to Google so it could spend its own effort on the layer it controls — the frameworks, the privacy story, the on-device integration — and it's asking developers to meet it there. The practical to-do list: invest in App Intents so your app is reachable from the new Siri; revisit Foundation Models now that it's multimodal and can run server-side; try the agentic Xcode workflow before reaching for an external tool; and budget for the regional and hardware gaps that will fragment who actually sees your AI features. The betas are live today — but the sessions, where Apple hides the API details and the server-side pricing, are where the real work begins.