Siri and Android Both Want Your App to Be an Agent Tool Now

Two announcements, two days apart, and they rhyme. On Tuesday Apple opened Siri AI to every app developer. On Thursday Google published how Android apps should expose their actions to on-device agents. Both are asking the same thing of app developers: describe what your app can do so that an assistant, not a person tapping through screens, can do it.

What Apple announced (Nov 3)

Alongside the iOS, iPadOS, macOS and visionOS 27.3 betas, Apple started a Siri AI Developer Preview. The mechanism is one Apple developers already know: App Intents. Any app can now bring its content and actions into Siri AI through intents, and Siri AI brings personal context, onscreen awareness and natural language on top.

Two details stood out to me:

• Skills. Developers can teach Siri AI to combine several of their actions into one experience, for requests too complex for a single intent.
• The Spotlight semantic index. Siri AI can surface any entity your app has put in that index, not only entities that fit one of Apple's predefined schemas. That loosens a real constraint: until now, the deepest Siri integration favoured apps that looked like Mail, Notes or Photos.

Apple's own apps (Calendar, Files, Mail, Maps, Messages, Photos, Reminders, Safari, Shortcuts) already use it. Customers get it "in a few months", and App Store submission opens "in the spring."

The limits matter as much as the feature:

• It needs an Apple Intelligence device set to a supported language: English, French, Japanese, Korean, Portuguese or Spanish to start.
• It's not available in the EU on iOS, iPadOS and watchOS, and not to users under 13.
• Server-side features have daily usage limits, and Apple says "expanded access to such features will be available for a fee in the future." That reads as a fee for users who want more, not a charge to developers, but Apple says terms and usage policies for Siri AI experiences come later.

What Google announced (Nov 5)

Google's write-up from the Android Developer Summit calls Android MCP its "unified tool-calling framework." It's built on the AppFunctions Jetpack library, and the name is deliberate: it's modelled on the Model Context Protocol, the same standard behind most coding-agent tools today.

Google lays out three ways for an agent to get something done in your app:

1. A remote MCP server you host, which gives you cross-platform reach.
2. Android MCP, running on the device. Google recommends it when the job involves end-to-end encrypted or sensitive local data, device sensors, local execution, or when there's no server at all.
3. If you've done neither, agents can fall back to Android's Computer Control framework and drive your UI directly, with no integration from you.

Around that, Google announced a Connected Apps Platform to reach users in the Gemini app and Search (a preview waitlist, "limited to selected production-ready apps in supported categories"), an Agentic API in ML Kit that uses function calling in Gemini Nano 4 to run agent loops entirely on-device, and a Jetpack Compose A2UI renderer in alpha that lets an agent stream native Compose UI from JSON.

Why the timing matters

Read them side by side and the shape is the same:

	Apple	Google
How your app describes an action	App Intents	AppFunctions (Android MCP), or a remote MCP server
Chaining actions	Skills	The agent orchestrates tools across apps
Your app's data	Spotlight semantic index	Local tools; remote MCP for server data
If you do nothing	Your app isn't reachable by Siri AI	Agents may drive your UI anyway via Computer Control
Status	Developer preview; ships in a few months	Available now; Connected Apps is a waitlist

That last row is the one I keep coming back to. On Android, skipping the integration doesn't keep agents out. It means they use your screens the way a person would, slower and more fragile, without you choosing what they can do. On iOS, skipping it means Siri AI simply can't reach your app. Different defaults, same conclusion: the actions you choose to expose are becoming part of your app's public surface, the way a web API is for a website.

For anyone shipping on both platforms, the practical point is to design the actions once. Decide which few things a user would ask an assistant to do in your app, give each a clear name, inputs and result, and then map that list to App Intents on iOS and AppFunctions on Android. The platform code differs; the thinking doesn't.

What I want to try

I haven't built against either preview yet. Here's my list for the coming weeks:

1. Install the 27.3 beta and Xcode 27.3, then read Apple's Siri AI Developer Preview docs for how skills are declared.
2. Write down the three or four actions in one of my apps that someone would plausibly ask Siri for, before writing any code.
3. Try the AppFunctions sample agent on Android with the same list, and see how much of the action design carries over.
4. Check what lands in my app's Spotlight index today, since Siri AI can now surface it.

Quick hits

• OpenAI says an agent pipeline built 948 Linux kernel root exploits (Nov 4). According to CSO Online's report on OpenAI's write-up, the pipeline worked against the kernel's master branch in mid-October, reproduced 1,724 vulnerabilities and produced 948 distinct privilege-escalation exploits for about $1.2M in API credits, with humans reviewing fixes before they went to maintainers. Expect more kernel patch churn in the base images your servers and CI use.
• GitHub Copilot is now inside Xcode's coding assistant (Nov 4). It's a public preview starting with Xcode 27.3 beta, available on every Copilot plan under Xcode > Settings > Intelligence. Mac and iOS developers now have another agent option without leaving Xcode.
• App Store subscription changes (Nov 2–4). App Store Connect can now show a message or offer when someone cancels a subscription, and a Retention Messaging API (on request) chooses which one each subscriber sees. Separately, the age-rating questionnaire gains "Shared Location" and "Dating and Relationships" questions from Nov 16, and every app must answer them by Jan 12, 2027.
• Gemini 2.5 models get redirected on Dec 1. Google's changelog says Gemini 2.5 Pro requests will go to Gemini 3.8 Flash and 2.5 Flash requests to 3.5 Flash-Lite, with breaking parameter changes (thinking_level replaces thinking_budget). If you pinned 2.5 in an app backend, test before December.
• Another npm hijack, with a twist (Nov 4). OX Security reports 17 legitimate npm packages were republished with malware that reads its command server address from the Ethereum blockchain, which makes it hard to take down. Pin versions, and consider a minimum package-age policy so brand-new releases wait a few days before you install them.

What I'm watching

.NET Conf, Nov 10–12, where .NET 11 is due to launch. That's next week's issue.

Sources

Siri AI and Android MCP

• Apple Developer, Siri AI Developer Preview announcement (Nov 3, 2026)
• Apple Developer documentation, Siri AI Developer Preview
• Android Developers Blog, What's new in intelligent apps at the Android Developer Summit (Nov 5, 2026)
• Android Developers, Android MCP / AppFunctions
• Daring Fireball, Apple begins opening Siri AI to developers (Nov 3, 2026)

Quick hits

• OpenAI, A thousand fewer paths to root
• CSO Online, OpenAI built an AI pipeline to find, exploit and patch Linux flaws (Nov 2026)
• GitHub Changelog, GitHub Copilot is now available in Xcode's coding assistant (Nov 4, 2026)
• Apple Developer, Retention messaging in App Store Connect (Nov 4, 2026)
• Apple Developer, Updated age rating questions (Nov 2, 2026)
• Google AI Studio, Gemini API changelog
• OX Security, PolinRider returns: hijacked npm packages use the Ethereum blockchain to conceal malware infrastructure (Nov 4, 2026)
• Microsoft, .NET Conf 2026