Your iPhone Is About to Control Every AI App You Use. Here's What This Means For You.
TL;DR
Apple’s real AI play is not a better chatbot — it’s making Siri the system-level gateway to AI across the iPhone. Nate argues WWDC’s biggest signal is a standalone Siri app that behaves like ChatGPT, while still letting Apple surface “ambient intelligence” from any app because it controls the full phone stack.
App Intents could turn the iPhone into an agent platform fast — especially for apps like Amazon, Uber, and photo editing. He says Apple is reportedly preparing intent-based integrations with major apps so users can ask Siri to do things like compare shoe prices or apply edits to photos without manually opening each app.
MCP support would be a major philosophical shift for Apple — and a huge unlock for tool-using agents. If Apple handles MCP at the system level, developers may not need to maintain their own protocol plumbing, and Apple could bring agentic tool access to a 1.5 billion-user install base that has mostly lived on ChatGPT Free-style experiences.
Apple appears to be splitting AI into private on-device tasks and outsourced cloud reasoning via Google Gemini. Nate expects Apple to run a small local model for sensitive data, then quietly hand off harder tasks like web research and deep reasoning to white-labeled Google models in the background.
The trade-off is clear: Apple wants reliable, controlled, single-task agents on iPhone — not the full open-ended “OpenClaw” workflow power users expect on desktop. He says Google is weaker than Anthropic or OpenAI on tool-calling harnesses, which suggests Apple is optimizing for personal consumer delegation on phones, while more complex long-running agents may live on devices like the Mac mini.
For builders, this is a call to design “agentic-first” apps now, before WWDC formalizes the framework. Nate’s advice is blunt: stop shipping deterministic apps with a thin chatbot layer, and start thinking about MCP, App Intents, and how your product behaves when Siri becomes the default AI front end.
The Breakdown
Apple Didn’t Lose the AI Race — It Picked a Different Battlefield
Nate opens by saying the popular take on Apple missing AI is too shallow. His real thesis is that Apple may have a neglected software advantage: a 1.5 billion-user iPhone install base, just as OpenAI is getting distracted by hardware rumors with Jony Ive, enterprise pressure from Anthropic, and the push to bundle ChatGPT and Codex into a super app.
Siri as a Standalone App — and an Invisible Layer Over Everything
Citing Bloomberg’s Mark Gurman, he says Siri is expected to become a standalone iPhone app with a ChatGPT-like interface, including multimedia conversations. But the bigger point is that Apple can make Siri accessible from anywhere on the phone, not just from one app — the “ambient intelligence” Greg Federighi hinted at when he said Siri should help you get things done wherever you are.
App Intents: The Builder Opportunity Hidden Inside WWDC
The next signal is Apple exposing agentic interfaces through something like App Intents, letting Siri pass clear requests into apps for remote interaction. Nate imagines demo partners like Amazon or Uber, then gives a very Apple example: asking Siri to apply a cinematic filter, clean up three faces, and crop someone out of a photo — because Apple already owns the world’s most-used camera.
MCP on iPhone Would Be a Huge Deal, Especially Coming From Apple
He lingers on MCP because Apple has historically hated outside connectors — “sometimes literal plugs,” as he jokes, nodding to the old connector wars and EU pressure. If Apple bakes MCP into the OS, then tool calling, compatibility, and security become system-level concerns, which would massively expand agent access for ordinary iPhone users instead of leaving it to power-user workarounds.
Why the Google Gemini Deal Matters More Than the Headline
Nate thinks Apple’s AI architecture will be split cleanly: a small on-device Apple model for privacy-sensitive work, and Google Gemini behind the scenes for harder reasoning, web access, and deep research. The catch is that Google hasn’t had the same momentum as Claude or OpenAI on long-running tool-calling harnesses, so Apple seems to be choosing dependable, consumer-grade delegation over full-blown multi-step professional agents on iPhone.
The Strategy: Protect the iPhone, Control the Interface, Fence the Ecosystem
Put together, he says Apple’s play has two layers: make Siri the default front door to AI, and re-open the app ecosystem in a tightly managed, Apple-approved way. He contrasts this with Apple’s recent anti–vibe coding posture — including its clash with Replit — and says Apple seems comfortable excluding a huge wave of looser builders if that’s what it takes to preserve security and the walled garden.
Why Apple Is Late — and Why This WWDC Actually Matters
He’s blunt that Apple telegraphed much of this at WWDC24, then failed to ship, leading to lawsuits, personnel changes, and a damaged Siri story. The issue now is whether Apple can return to its classic pattern — not first, but deeply integrated and seamless — while Google keeps shipping vision-based UI automation that already works, even if it’s more brittle.
Google, Samsung, and the Bigger Market Chessboard
For Google, Nate says the real prize isn’t just a billion-dollar deal but inference signal from iPhone users; it wants those complex mobile queries flowing to Gemini instead of OpenAI or Anthropic. He also brings in Samsung: if Apple can ship agentic features on a more affordable iPhone, it could undercut the Android pattern where the best AI features are reserved for $1,000-plus flagship phones.
What Builders — and Everyone Else — Should Do Now
His advice is practical: if you build apps, start designing for MCP and App Intents now, and ask whether your product is genuinely “agentic first” or just a normal app with a chatbot pasted on top. And if you’re not technical, his broader point is that the world is moving toward delegation, so people need to practice a new default: first ask whether an agent can do the work for you, then verify the result.