Agentic Android: Exposing and Consuming Capabilities with App Functions
Android 16 changes how the OS talks to installed apps. With the android.app.appfunctions package, Android moves away from the rigid deep links of App Actions and Slices toward an agentic model: apps are no longer just destinations to open, but headless toolboxes of capabilities that system agents can call directly.
This session goes past the documentation into a practical implementation of App Functions: wiring up an app to expose its features with the Jetpack androidx.appfunctions library, the @AppFunction annotation, and secure cross-process data with AppFunctionSerializable.
Then the other side: consuming those capabilities. I show a custom Android launcher that queries AppFunctionStaticMetadata and runs tasks from third-party apps through AppFunctionManager, turning an ordinary home screen into an AI-ready intent engine.
Key takeaways
The agentic OS shift
How the headless execution model of App Functions differs from legacy App Actions, architecturally.
Exposing capabilities
Implementing the androidx.appfunctions library, KSP code generation and the PlatformAppFunctionService entry point.
Schema and serialization
Defining structured parameters that system agents can easily index and call.
Live launcher demo
Querying the on-device index and running App Functions from a custom Android launcher.