The Gemini deal got the headlines. The developer migration problem didn’t.
Apple announced Siri AI at WWDC 2026, and the Bloomberg-reported licensing deal, approximately $1 billion per year for a model with approximately 1.2 trillion parameters, became the story. That’s understandable. A billion-dollar annual payment from Apple to Google for the engine powering Siri is a significant market signal, and the Markets and Regulation pillars both covered its implications.
What it crowded out: a concrete, time-sensitive engineering problem for any iOS or macOS developer who built voice functionality using SiriKit.
What SiriKit Deprecation Actually Means
SiriKit has been Apple’s voice integration framework since 2016. Developers used it to enable Siri interactions across specific intent domains: messaging (send a message via Slack), media playback (play this playlist in Spotify), ride booking, payments, and others. These integrations required implementing specific SiriKit intent classes, INSendMessageIntent, INPlayMediaIntent, and so on, and declaring them in the app’s Info.plist.
App Intents replaces this architecture. According to multiple developer-focused reports covering the WWDC announcement, SiriKit is reportedly being deprecated, and App Intents is now the mandatory framework for Siri AI integrations. The two frameworks are not backward-compatible in the way that allows silent migration. Existing SiriKit implementations don’t automatically become App Intents implementations. They require deliberate porting.
The specific migration scope, which intent domains, which APIs, which deprecated methods, requires verification against Apple’s WWDC developer materials. The deprecation claim itself comes from T3 and T4 developer-focused sources, not from the Apple Newsroom article body (which wasn’t fully readable in the source verification process). Don’t begin migration planning based on secondary reporting alone. Confirm the specific affected APIs in Apple’s developer documentation before scoping the work.
The Three-Tier Architecture Decision
App Intents isn’t just a framework replacement. It connects developer integrations to Apple’s new three-tier processing architecture for Siri AI, the architecture described in Apple’s WWDC materials and consistent with Apple’s existing Private Cloud Compute documentation.
The three tiers, as Apple describes them:
On-device, Apple Foundation Models Core handles requests locally on supported hardware. No data leaves the device. This is the privacy-optimal tier and the one Apple’s marketing emphasizes.
Private Cloud Compute (PCC), For queries that exceed on-device capacity, requests route to Apple’s Private Cloud Compute infrastructure. Apple’s PCC architecture is designed so that Apple itself cannot access the content of queries processed there. The PCC concept is documented and real, Apple’s security research team has published on it.
Google Cloud (stateless fallback), For queries that exceed PCC capacity, requests route to Google Cloud. Per Apple’s framing, this routing is stateless, no data is retained. The Google model involved is the one reported at approximately 1.2 trillion parameters, licensed for approximately $1 billion per year per Bloomberg.
For developers, the architectural question is: which tier will your users’ queries actually use? The answer depends on query complexity, device model, and the specific App Intent being invoked. A simple on-device task likely stays local. A complex multi-step request involving large-context reasoning may route to PCC or Google Cloud.
Who This Affects
Unanswered Questions
- Which specific SiriKit intent domains are deprecated, and on what timeline does Apple require migration?
- What is the routing logic by App Intent type, which query categories stay on-device, which route to PCC, which to Google Cloud?
- What is the confirmed Small Business Program download threshold for free cloud API access, per Apple's official developer documentation?
- Is the Extensions system (user-selectable AI model) available at WWDC launch or targeting iOS 27?
This routing logic matters for privacy disclosures, enterprise mobile policies, and regulated-industry deployments. If your app handles medical information, financial data, or anything covered by data residency requirements, you need to know, and be able to disclose accurately, where Siri AI requests from your app’s integrations actually go. Apple’s documentation should specify this routing logic per intent type; verify before updating your app’s privacy policy.
The Three-Tier Decision Tree for Developers
A working framework for evaluating your app’s position:
If your app handles sensitive personal, medical, financial, or regulated data, audit the routing logic before migration. Understand which of your App Intents queries will route to Google Cloud and whether that routing is acceptable under your privacy policy and applicable regulations.
If your app doesn’t handle sensitive data and your primary concern is capability, App Intents gives you access to the full Siri AI capability stack. The migration cost is real but the capability gain is significant: Siri AI connected to a 1.2 trillion parameter model is materially more capable than the Siri your users experienced before.
If you’re in the EU market, account for the delayed timeline. Apple’s EU rollout of Siri AI is reportedly delayed to fall 2026 for DMA compliance reasons. Feature planning for EU-market releases needs to build in that lag. This is T3-sourced; verify the specific timeline against Apple’s EU developer communications.
The Free Tier Question
Apple reportedly offers free cloud API access for developers in its Small Business Program who meet certain download thresholds. Multiple reports cite a figure under two million downloads as the qualifying threshold.
This specific figure hasn’t been confirmed against Apple’s official developer documentation. It appears consistently across T3 and T4 developer-focused sources, which gives it some credibility, but consistency among secondary sources isn’t the same as primary source confirmation.
The business implication is significant enough that it warrants primary source verification before any planning. If your app has under two million downloads and you qualify, cloud API costs for Siri AI integrations may be minimal. If you exceed the threshold or don’t qualify for the Small Business Program, you’re pricing a cloud API cost into your unit economics. Get to the Apple Developer portal and confirm the actual terms before building a cost model.
The Extensions System Timing
One reported capability that may not be immediately available: an Extensions system that would allow users to choose their preferred third-party AI model, Claude, Gemini, ChatGPT, as a Siri backend. MacRumors’ coverage references iOS 27 in this context, which suggests this feature may target a future OS release rather than the current WWDC announcement cycle.
If your development roadmap includes building for a user-selectable AI model backend via Siri, verify the availability timeline. Planning an iOS 26 feature launch around a capability that ships in iOS 27 is an easily avoided mistake.
Warning
Apple's privacy-first framing depends on architectural commitments Apple designed but Google's infrastructure executes at the cloud tier. The PCC documentation (security.apple.com) is the appropriate place to evaluate how robust those commitments are in practice. For enterprises handling regulated data, the question isn't whether Apple intends the stateless routing, it's whether your legal and compliance teams accept that architecture as sufficient for your use case.
The Strategic Context: What Google’s Role Means
The one structural observation that gets skipped in developer-focused WWDC coverage: Apple is licensing the inference layer of Siri from a company whose business model depends on data collection and advertising. Apple’s PCC architecture and the stateless Google Cloud routing claim are meaningful engineering commitments, and they’re testable. Apple’s security research team has published on PCC; independent researchers have audited aspects of it.
But the commercial structure is worth understanding clearly. Apple’s privacy-first positioning on Siri AI ultimately rests on a vendor relationship with Google and on architectural commitments that Apple has designed but Google’s infrastructure executes. The PCC documentation is the appropriate place to evaluate how robust those commitments are in practice.
What to Do Now
One: Confirm SiriKit deprecation scope and timeline against Apple’s developer documentation. Don’t migrate based on secondary reporting.
Two: Map your App Intents routing, which queries stay on-device, which route to PCC, which to Google Cloud, before updating privacy disclosures.
Three: Verify the Small Business Program threshold at the Apple Developer portal before building cloud API costs into your unit economics.
Four: If you’re building for EU markets, confirm the fall 2026 delay timeline against Apple’s EU developer communications and adjust feature planning accordingly.
Five: Wait on the Extensions system. If MacRumors’ iOS 27 framing is correct, building for user-selectable AI backends now is premature.