Source: Unsplash
Apple built a complete multi-AI switching system inside iOS 27 and then quietly left it out of the WWDC 2026 keynote. The Siri Extensions framework, discovered in the iOS 27 developer beta, lets users pick Claude, Gemini, or ChatGPT as the intelligence layer behind Siri, with a full settings panel and a dedicated App Store section already baked in. The feature is toggled off on Apple's backend, but the architecture is done. And for app designers, this changes everything about how we think about designing for the iPhone platform.
I've been building apps on Apple platforms for over eight years. Every major platform shift, the move to Retina, the jump to SwiftUI, the introduction of Dynamic Island, each one forced a rethink of certain design assumptions. But this one is different. This isn't a screen-size change or a new input method. This is Apple quietly dismantling the idea that every iPhone user experiences the same AI brain, and replacing it with a user-chosen intelligence layer that sits between the person and the app.
That is a foundational shift. And most designers I talk to aren't fully tracking it yet.
What Apple Actually Built Inside iOS 27
Here's what the iOS 27 beta reveals. Apple has built a framework called Siri Extensions that creates a standardized integration layer for third-party AI providers. Developers behind Claude, Gemini, and ChatGPT can plug into this framework through their App Store releases. When an iPhone user asks Siri something, Siri no longer has to answer with Apple's own model. It can route the query to whichever AI the user has set as their default.
The implementation details matter here. Siri remains the orchestration layer. It handles the system-level stuff: on-device context, app integrations via App Intents, privacy controls. But the reasoning, the generation, the conversational intelligence, that can now come from a third-party model the user has chosen. The settings panel is already built. The App Store section for Extensions already exists. Apple just hasn't flipped the switch yet.
"Siri Extensions may become Apple's next major developer battleground, because developers will compete for a place inside the assistant layer rather than only for home-screen placement or App Store ranking."
— AppleMagazine, June 2026
That quote lands differently when you think about it from a product design perspective. The App Store made app icons the primary currency of mobile real estate. Then notifications took over for re-engagement. Then widgets. Now Apple is building the infrastructure for a new tier: assistant-layer integration. And unlike widgets or notifications, this one is driven by natural language, meaning the quality of your AI integration, not just your icon design or push strategy, determines how often users reach your product through Siri.
Why This Matters More Than Anything Else at WWDC 2026
iOS 27 introduced a lot of things. Siri AI got smarter. Apple Intelligence expanded. The Foundation Models framework let developers run AI inference on-device. Those are all significant. But the hidden Siri Extensions system is the most structurally important thing Apple has built this cycle, precisely because it wasn't shown in the keynote.
Apple spent years fighting to keep the iPhone experience uniform and controlled. App Store guidelines, design language standards, hardware optimization, all of it pointed toward a single coherent experience that Apple curated. The moment Apple allows users to plug in Anthropic or Google as the intelligence layer behind Siri, that coherence changes. Users will start experiencing meaningfully different AI behaviors on what looks like identical hardware, and app designers will have to build for that variability.
Via GIPHY
Think about what this means concretely. If your app uses App Intents to expose actions to Siri, the user's chosen AI model is now the one parsing their intent and deciding how to fulfill it. Claude interprets natural language differently than Gemini. Gemini behaves differently than ChatGPT. Your App Intents schema will be the interface between your product and whatever model the user has set as their default. If your intents are ambiguous, different AI models will resolve them differently. You now have to design for that.
The Four Design Shifts App Teams Need to Make Before iOS 27 Ships
I've been thinking through what this actually requires from a product design standpoint. Here's where I'd focus if I were building for iOS 27 today:
- Audit your App Intents for model-agnostic clarity. Your App Intents are going to be called by different AI models with different parsing behaviors. An intent that's crystal clear to ChatGPT might get misinterpreted by Claude because of how the parameters are named or how the response types are structured. Go through every intent you expose and ask: is this unambiguous enough that any LLM would resolve it the same way? If not, fix it now. This is not optional.
- Design explicit handoff moments. When Siri hands a query off to a third-party model, users need to know it happened. Apple's own interface includes voice differentiation, meaning the voice changes depending on which AI is responding. But in-app, you'll need your own handoff signals. If your app is getting invoked through a Gemini-powered Siri call, your UI needs to acknowledge that context without confusing the user. This is new UX territory and there are no established patterns for it yet.
- Stop treating on-device AI as a secondary fallback. Apple's Foundation Models framework puts real inference capability directly on device, no API call required. The smart product design move is to use on-device Foundation Models for latency-sensitive, context-aware operations, and reserve third-party Siri extensions for complex reasoning. Build a two-tier AI architecture in your app now, before iOS 27 ships, so you have the flexibility to slot in the right model at the right moment.
- Prepare for user trust to transfer through the AI, not just through your brand. Here's the one that's hardest to operationalize: if a user loves Claude and trusts it deeply, and your app gets invoked through Claude via Siri, that trust partially transfers to the experience of your app. The opposite is also true. If the AI model a user chose handles your App Intent poorly, the negative experience lands on your product even though you didn't build the model. Your app's reputation is now partially dependent on the quality of third-party AI integrations you have no direct control over. That's a new kind of design risk that requires new kinds of thinking.
SiriKit Is on a Deprecation Clock. What That Means in Practice
One thing that came through clearly in the iOS 27 developer documentation is that SiriKit is on a deprecation path. App Intents is now the mandatory way Siri communicates with apps. If your product is still built on legacy SiriKit integrations, you're working off a shrinking runway.
The migration from SiriKit to App Intents is not trivial. The paradigm is different. SiriKit was about mapping user speech to predefined domains (messaging, payments, etc.). App Intents is about exposing arbitrary actions your app can perform, with structured parameters, and letting Siri and now other AI models figure out how to invoke them based on natural language. The flexibility is much greater. The design responsibility is also much greater, because you're now responsible for defining a vocabulary of actions that AI models will query against.
In practice, this means your product design team needs to collaborate much more closely with engineering on the App Intents schema. Historically, App Intents were an engineering task that design signed off on at the end. That won't work anymore. The naming of parameters, the structure of responses, the way you handle ambiguous inputs: these are design decisions that directly affect how well your product works inside an AI-orchestrated environment. Designers need to own this layer of the product.
The Bigger Picture: Apple Just Opened the iPhone to an AI Ecosystem
I want to zoom out for a second because I think it's easy to get lost in the implementation details and miss the strategic significance of what Apple has built.
For over a decade, Siri was Apple's answer to the question "who should be the intelligence layer of the iPhone." The answer was always: Apple. Now, for the first time, Apple is building infrastructure that lets the answer be Google, or Anthropic, or OpenAI. That's an extraordinary reversal of platform philosophy, and Apple did it quietly, without a keynote announcement, which tells me they're still working out the business implications.
But the architecture is real. The beta code is shipping to developers right now. And the competitive dynamic this creates is fascinating. Anthropic and Google are both suddenly in a race to be the default AI on the most valuable personal device on the planet. Not through their own apps. Through Siri.
For product designers, the lesson is: the platform layer is no longer fixed. You used to design for "the iPhone experience." Now you have to design for "the iPhone experience, filtered through whichever AI the user has chosen as their default assistant." That's a more complex brief, but it's also a more interesting one. The designers who lean into that complexity now will be the ones building the most compelling experiences when iOS 27 ships this fall.
I've been writing about AI-native product design challenges for a while now at Medium and the reloadux blog. The iOS 27 Siri Extensions story is exactly the kind of platform shift that doesn't announce itself loudly. It shows up in a developer beta, quietly, while everyone's watching the keynote. And then six months later, it's the thing everyone is scrambling to redesign for.
Don't be the team that scrambles in November. Start now.
Are you building for iOS 27 already? Have you started migrating your App Intents schema with multi-AI in mind? I'd love to hear what you're finding in the trenches. Drop a comment below and let's compare notes.
Sources: 9to5Mac (9to5mac.com), MacRumors (macrumors.com), AppleMagazine (applemagazine.com), Tom's Guide WWDC 2026 Recap (tomsguide.com), The Next Web on Apple Siri Extensions (thenextweb.com), TechCrunch WWDC 2026 Recap (techcrunch.com), Medium: iOS 27 for Developers by Alok Upadhyay (medium.com/macoclock), Analytics Insight: iOS 27 Hidden Features (analyticsinsight.net), Let's Data Science: Apple Hides Siri Extensions (letsdatascience.com), Beam AI: WWDC 2026 Agent Extensions (beam.ai)