WWDC 2026 for AI Agent Builders: App Intents Goes Mandatory, MCP Goes System-Wide, and Siri Gets a Gemini Brain

What It Is / 这是什么

At WWDC 2026 (The Verge), Apple dropped the biggest platform shift for AI developers since Shortcuts launched in 2018. Three moves matter:

1. App Intents is now mandatory. Every app on iOS 27 and macOS 27 must declare App Intents — structured, machine-readable descriptions of what the app can do. No intents, no Siri integration, no Spotlight agent actions.
2. MCP goes system-wide. Apple adopted Anthropic's Model Context Protocol as the OS-level standard for tool discovery and invocation. Any LLM — Apple's on-device models, Private Cloud Compute, or your own provider via the new LanguageModel Swift protocol — can call App Intents through MCP.
3. Siri is now an agent. Siri can chain multi-step, cross-app tasks. The cloud-side orchestration layer runs on Gemini through a Google partnership, while Apple's own on-device foundation models handle privacy-sensitive work (AppleInsider).

Why It Matters / 为什么重要

Before WWDC 2026, building an AI agent that could actually do things on an iPhone meant fragile screen-scraping or begging for a SiriKit domain. Now, Apple has built the rails:

• 1.5 billion devices become agent endpoints. Every app that declares App Intents is instantly discoverable by every LLM on the system. Your app becomes a tool.
• MCP eliminates the integration tax. One protocol. Every model. Every app. No per-LLM adapter code.
• Siri's agentic upgrade is real. Stork.ai's WWDC grade report (Stork.ai) calls the new Siri "the first time Apple shipped a credible agent" — it can book reservations, modify calendar events across apps, and compose messages with context from your email.

Key Features & Specs / 核心特性

How It Compares / 横向对比

vs. Android + Gemini. Google made Gemini the system agent on Android in 2025, but relies on Accessibility Services for app interaction — a hack. Apple went deeper: App Intents give structured, typed tool access. The tradeoff is developer effort — every app must declare intents. Android's approach works with zero developer action but is brittle.

vs. OpenAI operator / computer use. Cloud-based computer-use agents need screen pixels and mouse events. Apple's MCP + App Intents approach gives agents semantic access to app functionality — faster, more reliable, and privacy-preserving since the LLM never sees your screen.

vs. Anthropic's MCP ecosystem. MCP on Apple's OS is the same protocol Anthropic shipped in late 2024. The difference: Apple bakes it into the hypervisor layer. Claude on a Mac can now call your app's intents through the same MCP server that Claude Desktop uses.

Who Should Use It / 谁应该关注

• App developers: Start declaring App Intents now. In iOS 27, intents are your app's API surface to every AI agent on the planet.
• Agent framework builders: The Foundation Models framework's LanguageModel protocol is the new integration point. Wrap your LLM in 2 protocols and every Apple device becomes a distribution channel.
• Enterprise mobility teams: MCP-based tool calling on-device means your internal tools can be agent-accessible without sending data to a third-party cloud.
• Indie developers: Siri is now a real agent with cross-app reasoning. If your app solves a narrow task well and exposes it as an intent, Siri will find it and route users to you.

FAQ / 常见问题

See structured FAQ in the frontmatter above. Key takeaway: the deadline is September 2026 public launch. If your app doesn't declare App Intents by then, it's invisible to the agent layer. Start porting now.

Sources: Apple Developer Session 339 · The Verge WWDC 2026 · AppleInsider Foundation Models · Stork.ai WWDC Grades