From Suggestions to Actions: How Android Opens the App Layer

The Android interoperability measures become most consequential when an assistant stops merely answering and starts changing state in another application or in the operating system. This chapter separates three action mechanisms in the Commission's design: structured app integrations, screen automation, and system integration. That comparison gives readers a practical test for authority: what has an app explicitly exposed, what can an assistant imitate, what can it change in Android itself, and which controls keep those powers visible and bounded.

The distinction matters because “cross-app action” can describe very different things. An app may publish a defined function such as creating a calendar event. An assistant may instead inspect another app's screen and imitate taps and typing. Or it may call an operating-system capability that changes Bluetooth, media playback, or a device setting. These paths do not have the same reliability, observability, or failure modes, even when they help complete the same user request.

Structured integrations: actions apps choose to expose

Structured on-device integration is the most explicit of the three mechanisms. An application exposes functions that another service can discover and invoke. The assistant needs a way to find what is available, retrieve the metadata needed to choose the right function, execute it, and let the user know that the execution is taking place. Where the integration is designed to run inside the app, the task can be completed without an unnecessary switch to another screen.

The difference from general screen control is the contract. A structured function describes an action in terms the system and the participating apps can recognise: send a message, create a note, schedule a meeting, retrieve an email, or start navigation. The app is not handing over its whole interface. It is making particular capabilities available through a common channel. An assistant can therefore reason about an action's meaning instead of inferring it only from pixels, labels, and the current arrangement of buttons.

That contract does not remove the user's role. The final measures specify that the user should be aware of and able to observe a structured integration while it is being executed. The app and the user determine which tasks are made available to AI services. The assistant can discover exposed functions, but discovery is not the same as permission to perform every possible operation in the app. A shopping application might expose the ability to add an item to a list while withholding account changes or a purchase confirmation. The exact boundary belongs to the app's integration choices and the applicable Android controls.

The requirement is also intended to prevent Google's own channels from remaining uniquely capable. The measures name Android's existing integration channels, including App Actions and App Functions, and require third parties to receive access in a way that is equally effective to Alphabet's services. For Qualified Services, the specified scope reaches into selected Google applications. An assistant may, for example, retrieve or draft mail, manage a calendar event, start navigation, control media, work with permitted messages, or initiate a phone call. The point is not the length of that list.