The Bottleneck

Every WWDC preview analysis is telling the same story right now: Apple is late to the AI party but gets to pick the music. The short version: Apple isn't building a chatbot. Apple is building the runtime every agent has to pass through. MCP wired directly into the operating system. 1.5 billion iPhones as a distribution channel. Platform play, not model race.

The analysis is correct. And the cleverest part is a detail that gets lost in the enthusiasm.

Apple is implementing MCP openly. Any agent – ChatGPT, Claude, Gemini, local open-source models – can access app capabilities through the protocol. No vendor lock-in to Siri. That sounds generous. It's the opposite.

Because access to the protocol isn't the control point. The control point is the gateway underneath: App Intents. Apple's framework decides which actions an app makes visible to agents in the first place. Apps that don't implement intents become invisible to agents – regardless of which agent, regardless of which model. Nate puts it himself: "The toll booth just moved from 'you need our app store to distribute' to 'you need our intent system to be actionable.'"

This is cleverer than closing the door. Closing a protocol provokes resistance, regulation, antitrust lawsuits. Opening the protocol while controlling the gateway gets you cooperation. Every agent developer using MCP strengthens the Apple ecosystem. Every app developer implementing intents makes the platform more valuable. Apple doesn't have to force anyone. The architecture creates the pull on its own.

We know this pattern one layer down. The App Store: open submission, closed approval. 30% commission. Guidelines that change when it suits Apple. Technically open, structurally controlled. Sold as quality assurance. The agent runtime is the same principle, one abstraction higher.

But there's a second asymmetry that nobody is working through publicly. Apple has Google under contract to power Siri's cloud reasoning with Gemini. Apple pays an estimated one to five billion dollars per year. Google delivers the intelligence. Apple keeps user data on-device; Google never sees it.

What Google does see: the query patterns of 1.5 billion iPhone users. Which tasks they delegate. Which workflows fail. Which intents get triggered most frequently. The raw data stays private. The aggregate is gold. Apple is buying intelligence while giving away information that makes Google's Android agent better. Apple knows this. Nate explicitly calls the deal "temporary." The question is how much Google learns in the meantime.

For dekodiert readers, these are two layers of the same topic: Who controls the gateway through which AI becomes actionable? And who benefits from the information flows that emerge in the process – even when they're not the raw data?

Mattes Schrader says on synthszr.com: "Apple has never lost a platform game." But this game has a new player: the intelligence supplier who is simultaneously the strongest competitor.