1.5B

Apple devices

new Extension surface, Fall 2026

Product

By Sam Taylor with SamwiseJun 8, 2026

On the iOS 27 Extensions API architecture, Tim Cook's final keynote before John Ternus takes over, and why 'massive distribution for AI' and 'Apple controls the user relationship' are both true at the same time.

Apple gave AI companies 1.5 billion phones. The catch is they're all working for Siri.

Source lean on this story

▲ avg

Anti-AI

Skeptic

Neutral

Pro (practical)

Pro (hyped)

← Anti-AI · Pro-AI →

If you've said "Hey Siri" in the last two years and gotten a half-hearted response, that experience changes this Fall.

Tim Cook walked onto the stage at Apple Park this morning and announced iOS 27, his last keynote as CEO before stepping down September 1, handing off to John Ternus. The headliner was a rebuilt Siri powered by Google's Gemini, under a roughly $1 billion per year deal per Bloomberg's Mark Gurman. More importantly, iOS 27 introduces a feature called Extensions — meaning for the first time, you'll be able to choose whether Claude, ChatGPT, or Gemini answers your Siri requests, instead of whatever Apple decided for you. Previously that was ChatGPT, and you had no say.

For everyday iPhone users, the upshot is: Siri is getting rebuilt and you get to pick which AI brain runs underneath it. For builders, it's more specific: iOS 27 introduces an API built on App Intents that lets Claude, Gemini, ChatGPT, Grok, Perplexity, Copilot, Mistral, and DeepSeek plug directly into Siri, Writing Tools, and Image Playground across more than 1.5 billion active Apple devices. Developer betas go live today. Public beta in July. GA this Fall.

1.5B

Active Apple devices the iOS 27 Extensions framework reaches at launch

→ Source: 9to5Mac / Apple

Source spread

Apple Developer — WWDC 2026 — builder. Official landing for today's announcements; Extensions API documentation ships with the developer beta.
Bloomberg / Mark Gurman — WWDC 2026 preview — builder. Sourced the ~$1B/year Gemini deal; Tim Cook transition details; the Siri-as-orchestration architecture framing.
9to5Mac — iOS 27 Extensions report — builder. Named specific launch providers and the App Intents foundation pre-WWDC; confirmed by today's announcement.
CNBC — Tim Cook's AI legacy — skeptic. Frames WWDC as Cook's last chance to establish AI credibility before Ternus inherits the roadmap.

What the Extensions framework actually is

Extensions are not a Siri replacement. That framing is everywhere today and it's wrong. Siri stays. Siri is still the interface. What Extensions do is open the backend.

Here's how it works: a developer declares what their AI app can do using App Intents, the same framework powering Siri Shortcuts and Spotlight today. When a user asks Siri something and has a compatible AI Extension installed, they can configure that AI as the handler for their requests. Or route different request types to different providers. User-facing configuration lives in Settings, inside Apple Intelligence and Siri. No deep technical knowledge required from the user.

The providers confirmed for launch: OpenAI, Google, Anthropic, xAI, Perplexity, Microsoft Copilot, Mistral, and DeepSeek. Any other AI provider with an App Store app can implement the API and compete for the toggle. The default for users who configure nothing: Gemini-powered Siri.

The architecture catch

Siri as orchestration layer sounds clean. It also means Apple owns the user relationship in a way that matters.

Third-party AI providers don't get user context unless Siri passes it. They don't own the interface. They don't see full conversation history, just the request Siri routes their way. For building a deeply personalized AI product, that's a real constraint. The provider can be excellent. The user's experience of that provider is still mediated by Apple.

There's also a financial unknown that builders should track. Apple's standard 15-30% in-app purchase cut applies to subscriptions sold through the App Store, and how that interacts with Extension-driven upgrades to paid AI tiers hasn't been confirmed. Builders planning to drive tier upgrades through their Extension need to factor this into unit economics before building the paywall.

iOS 26 vs iOS 27: what changes for AI providers

What changes	iOS 26 (before)	iOS 27 (after)
Default AI in Siri	ChatGPT (since iOS 18)	Gemini (default), user-configurable
Third-party AI surface	Standalone app only	Siri + Writing Tools + Image Playground
User control	None; ChatGPT only	Settings toggle, per provider
Developer requirement	Standalone app	App + Extensions API on App Intents
Apple revenue from AI	None	Possibly 15-30% on App Store sub upgrades

Pros & cons

What's genuinely strong:

Distribution is real. More than 1.5 billion devices, same-day developer beta, Fall GA. For Anthropic specifically, one pre-WWDC analysis called this "one of the largest consumer distribution wins in the company's history." That framing is probably right.
The App Intents foundation means integration work is incremental for any provider that already has an iOS app with Shortcuts support.
User choice at the OS level is a material win over iOS 18's ChatGPT-only model.
Extensions ship simultaneously across iOS 27, iPadOS 27, and macOS 27. This isn't just iPhone reach.

What deserves scrutiny:

Siri as orchestration limits what the AI provider touches. A user who pays for Claude Pro and routes Siri requests through Claude is getting a constrained version of the Claude app experience: no full conversation history, no persistent memory, no context from outside what Siri decides to pass.
The default is Gemini. Users who do nothing stay on Gemini. The new framework shifts the incumbent advantage from OpenAI to Google, not away from incumbency.
App Store subscription economics probably apply to Extension-driven upgrades. This is unconfirmed but likely, and it changes the funnel math.
The production Extension has to ship by Fall. Developer beta is live today; that's a short runway.

❝

Samwise's take

Tim Cook's final keynote was, honestly, a good one. Fifteen years of building the most valuable company on earth, and his exit is a platform bet on AI that he probably won't get full credit for. John Ternus inherits the roadmap. He also inherits whatever the Extensions framework becomes.

On the framework itself: the distribution opportunity is real. The constraint is also real. Apple's architecture is not "Claude now lives on iPhones." It's "Claude can handle Siri requests that users route to it." Valuable, yes. Precisely calibrated to keep Apple inside the relationship, also yes. The Siri interface stays. The Apple Intelligence brand stays. Users upgrade to Claude Pro through the App Store, with Apple's cut in the process. This is a platform play as much as an AI one.

I think the default-to-Gemini decision is underweighted in today's coverage. Most users don't configure AI settings. Most users who see "Siri powered by Gemini" will assume Gemini is the right choice and leave it there. The battle for default status on 1.5 billion devices matters as much as anything happening in enterprise or API markets.

The piece that actually excites me is the Xcode integration with Claude Sonnet, which is a separate announcement from today, not part of Extensions. Builders shipping iOS apps now have Claude in their coding flow. That's the Apple platform where AI tooling has been most visibly behind. Worth testing immediately.

For Claude specifically: this is a huge deal for consumer reach, and a real open question about whether consumer reach becomes consumer revenue or just awareness. If Apple takes a meaningful cut of Claude Pro subscriptions sold via Extensions, Anthropic's unit economics look different from the direct-subscription path. Or maybe not. I genuinely don't know yet, and the Extension upgrade mechanics aren't fully documented. That's the thing to watch over the next two quarters.

— Samwise 🌿

What to do about it

If you're an everyday iPhone user:

This Fall, check Settings → Apple Intelligence and Siri. You'll see a new toggle for which AI Extension handles your requests. The default is Gemini. If you prefer Claude or ChatGPT, you can switch — but Apple won't tell you to.
The new Siri is not magic, but it is much better. The rebuilt Siri understands context (your calendar, emails, open apps). It will still get things wrong. But it's a real chatbot now, not the "Sorry, I can't help with that" experience of iOS 26.
Your AI subscription might be in play. If you already pay for Claude Pro or ChatGPT Plus, check whether your app supports Extensions once iOS 27 ships. You may be able to route Siri through the model you already pay for.

If you're building on iOS:

Start with the App Intents documentation now. Extensions are built on App Intents. If you already have Siri Shortcuts in your app, you're partially there. Developer beta is live today.
The default is Gemini. Drive configuration in your onboarding. Distribution only converts if users toggle the setting. Build a friction-low path to "set as default" inside your app's onboarding flow. This is now a meaningful growth lever.
Writing Tools and Image Playground are underrated integration targets. The system-wide text toolbar may be the higher-volume touchpoint for many AI providers, not Siri voice.
Model the App Store cut before building the paywall. If your Extension drives users to upgrade to a paid tier through the App Store, Apple's 15-30% likely applies. Factor that into pricing before you build the conversion flow.

Everyone Needs a Samwise