--- title: "Agent-ready apps: AppFunctions, App Intents, and why AI assistants will use your app" description: "AI assistants are learning to use apps for us. How AppFunctions on Android and App Intents on iOS work, and how to make your app agent-ready." author: "MH TOUFIK" published: 2026-11-06T03:00:00.000Z updated: 2026-11-05T08:02:49.000Z url: https://toufik.dev/blog/agent-ready-apps-appfunctions-app-intents category: "Mobile AI" tags: ["AI agents", "Android", "App Intents", "AppFunctions", "iOS"] --- # Agent-ready apps: AppFunctions, App Intents, and why AI assistants will use your app For fifteen years, using an app has meant the same thing: find the icon, open it, tap through the screens. That's starting to change. On both Android and iOS, **AI assistants are learning to use apps on our behalf**. You say "book my usual ride home" and the assistant does it, without you opening the ride app at all. If you build apps, this raises a new question: **can an assistant actually use yours?** ## How it works ![How an AI assistant uses your app](https://toufik.dev/media/diagram-03-agent-ready-apps-v2-38643e89.webp "You ask in plain words; the assistant finds a function your app declared, your code runs on the phone, and the assistant confirms the result.") The idea is the same on both platforms: 1. You **declare** what your app can do, as clear, named functions with inputs. 2. The system's assistant **discovers** those functions. 3. When a user asks for something in plain words, the assistant **picks the right function** and fills in the inputs. 4. **Your own code runs** on the phone, with your rules and your checks. 5. The assistant **shows the result**, and asks before anything sensitive. ## On Android: AppFunctions Google introduced **AppFunctions** in February 2026, and at Android Dev Summit in October it moved to **beta**. In Google's words, AppFunctions lets apps "expose data and functionality directly to AI agents and assistants". You use the AppFunctions Jetpack library to create self-describing functions that agents can discover and run from natural language. Google compares it to MCP, the protocol many AI tools use to call server-side tools. The difference is that **AppFunctions runs on the device**, not on a server. That's good for privacy and for speed. A few details from Google's announcement worth knowing: - Users can **watch progress** in a notification or "live view" and **take over manually** at any point. - Gemini is designed to **ask before sensitive tasks**, such as making a purchase. - Samsung Gallery on the Galaxy S26 series was one of the first apps to support it. Google is also testing a **fallback**: on some Galaxy S26 and Pixel 10 phones, Gemini can work through an app's screens for you, even if the app declares no functions. It started as a beta in the US and Korea for food delivery, grocery and rideshare apps. It's clever, but it's slower and more fragile than a declared function. **Declaring functions keeps you in control.** ## On iPhone: App Intents Apple has had a version of this for years: **App Intents**. You describe your app's actions in Swift, and the system can show them in Siri, Shortcuts, Spotlight and widgets. Apple keeps building its newer intelligence features on top of the same framework. If you already have App Intents for Shortcuts, you're ahead. Many of those same intents become the actions an assistant can call. ## Why this matters for your app A few reasons I think every app team should pay attention: - **Discovery changes.** Today users find apps in a store. Tomorrow, an assistant may pick the app that can *actually do the task*. Apps that expose nothing get skipped. - **Daily actions become effortless.** The small repeated jobs, like reordering, checking a balance or adding a reminder, are exactly what people will hand to an assistant. - **It's a better experience on small phones.** Saying one sentence beats five taps on a slow device. I'll be honest: nobody knows how fast this will grow. Some commentators think assistants might replace many apps; others think it's a short-term step. But both platforms are investing heavily, and the first move is cheap. ## How to get started Here's how I'd approach it for an existing app: 1. **List the 3–5 things users do most often.** Not every feature, just the ones people repeat every day or every week. 2. **Write each as a clear function:** a name, simple inputs and a clear result. For example `requestRide(destination)`, `addToCart(item, quantity)` or `getExamResult(roll, exam)`. 3. **Keep your rules in your code.** Validation, permissions and limits should live in your function, not depend on the assistant behaving well. 4. **Make sensitive actions ask first.** Payments, deletions and messages to other people should always need the user's confirmation. 5. **Ship it on one platform first.** App Intents on iOS is mature today. AppFunctions on Android is in beta, so it's a good time to experiment. ## My take We spent years designing apps for thumbs. The next few years add a second user: the assistant. The good news is that the work is mostly what good app design already asks for. Know what your users really come to do, make those actions clear and reliable, and keep your rules in your own code. Do that, and your app will be ready whether people tap it or talk to it.