Apple shipped the announcement everyone expected. Siri AI is real, Gemini is running it, and the developer surface is App Intents.
At its WWDC 2026 keynote on June 8, Apple confirmed that Siri AI is powered by Google’s Gemini models, turning a licensing deal first reported at roughly $1 billion per year into a live product feature. The integration marks the first time a third-party frontier model runs inside Apple’s flagship assistant at the OS level. What that means in practice is a more capable Siri, but the capability set is specific, and some details are still beta-stage promises.
What’s confirmed
On-screen awareness is real and developer-documented. Apple’s own developer documentation, “Making onscreen content available to Siri and Apple Intelligence”, confirms that Siri AI can respond to questions and take action based on onscreen content. A user can ask about what’s on their screen without copying text or switching apps. That’s a structural change from previous Siri behavior, where context was almost entirely voice-input-dependent.
Cross-app synthesis is confirmed in Apple’s own product language. Apple’s Apple Intelligence page describes requests like “Send the email I drafted to April and Lilly”, Siri AI understanding which draft email to pull, which contacts to address, and executing across apps without the user specifying each step. The capability is framed as cross-app action, not just cross-app lookup.
Unanswered Questions
- Which apps have already shipped App Intents compatibility, and what's the realistic coverage for a typical knowledge worker's stack?
- What are the latency and accuracy characteristics of cross-app synthesis requests in beta, Apple has not disclosed these.
- How does Siri AI handle requests that span apps whose developers haven't implemented App Intents?
The developer path
App Intents is the integration framework. The developer.apple.com WWDC 2026 portal lists App Intents in its Technologies section, meaning developers who want their apps to participate in Siri AI’s cross-app synthesis need to implement App Intents. That’s not a small lift for teams that haven’t started. The framework has existed since iOS 16, but the stakes for implementing it just changed.
What’s still reported, not confirmed from primary specs
The Liquid Glass visual interface and Dynamic Island integration are confirmed in broad WWDC 2026 coverage, according to that coverage, Siri AI appears from the Dynamic Island with Apple’s Liquid Glass visual language on supported iPhones, but these details aren’t enumerated in the developer portal’s fetched content. Device compatibility reportedly extends to the iPhone 11 lineup, though Apple hasn’t published final compatibility specifications. The initial rollout is designated as a beta, with access reportedly managed via waitlist.
What the Gemini layer actually means
Gemini runs the reasoning. Apple’s on-device models handle the privacy layer and context parsing. The architecture follows a pattern visible across the industry this week, on-device intelligence for the sensitive work, cloud frontier models for the complex work. Siri AI doesn’t replace Apple Intelligence; it sits on top of it, handling requests the on-device models can’t resolve.
The catch is that none of the performance benchmarks were disclosed. Apple hasn’t published latency figures, accuracy rates on cross-app task completion, or how Siri AI performs when the request spans more than two apps. Production teams deploying enterprise iPhone fleets should treat the beta framing literally, real capability data won’t exist until independent evaluations run on the GA release.
Siri: Before and After WWDC 2026
What to watch
App Intents adoption rate among third-party developers will determine how broadly the cross-app synthesis feature actually works. An assistant that can bridge Notes and Mail is useful; one that bridges the 40 apps a knowledge worker uses daily requires those apps to implement the framework. Expect the developer story to unfold over the next iOS 27 beta cycle. Final device compatibility specifications and the waitlist-to-GA timeline are the other near-term signals.
Don’t expect this to replace dedicated AI tools in professional workflows yet. Siri AI’s confirmed capabilities are impressive for a system-level assistant. They’re not a replacement for purpose-built coding, writing, or research tools. The value is in ambient assistance, the tasks users already attempt with Siri but abandon because it fails. That’s a real improvement. It’s just a narrower one than the keynote framing suggests.