Google Expands Gemini on Android and Pixel With App-Running AI Feature

Your phone is about to start doing the tapping for you. Google has previewed a new Gemini feature that can execute multi-step tasks inside Android apps, from booking rides to reordering groceries and food.

In a Google announcement, the company said the beta will let users trigger Gemini to carry out supported actions in the background while providing live notifications so they can monitor progress or step in at any time.

A beta rollout with tight boundaries

The feature is launching as a beta inside the Gemini app, initially on Pixel 10, Pixel 10 Pro, and Pixel 10 Pro XL devices, as well as Samsung’s Galaxy S26 series, with availability beginning in the US and Korea. 

Users activate it by long-pressing the power button and asking Gemini to complete a supported task, which then runs in the background while they continue using their phone.

At launch, the automation is limited to select apps in food delivery, grocery, and rideshare categories, reflecting a controlled early phase. Compatibility and availability may vary; the feature is restricted to certain apps, and it is intended for users 18 and older as testing begins.

Stepping into the on-device agent era

Digital assistants have long stayed in advisory mode — answering, suggesting, and reminding. With this update, Gemini is being pushed into execution, completing tasks within apps rather than returning instructions to the user.

That change matters in a crowded field where tech companies are racing to build assistants that can actually carry out work. By keeping the feature on-device and limiting it to supervised, user-triggered commands, Google is testing how far it can let software handle real-world actions on a phone without handing over full control.

Live visibility into every step

Google is going out of its way to stress oversight. Gemini is not being introduced as software that takes control in the background. Instead, users are meant to see each step as it happens and retain the ability to intervene before anything is finalized.

That emphasis comes as task-running software moves from demos into everyday use. When a system can place orders or book services inside third-party apps, mistakes carry real consequences. Requiring a direct command to begin and surfacing progress in real time limits how much authority the feature has from the start.

Access is tightly constrained as well. Gemini runs in a contained environment on the device, within supported apps only, without broad system permissions. The structure keeps it closer to a supervised shortcut than an independent operator.

As companies compete to make software more active inside consumer devices, Google’s rollout leans toward caution. Expansion may follow, but the early rollout focuses on supervision and user control.

Google is expanding Gemini’s creative side as well, sharing guidance on getting better results from its Lyria 3 music generator.