1. News
  2. Software & developer tools
  3. Google · Android AI
Google · Android AI2 min read

Google lets AI assistants carry out tasks across Android apps

Google is building tools so Gemini can finish multi-step jobs across your Android apps, with some of the work staying on your phone.

By the Dashline deskSource: Android Developers blog
Google lets AI assistants carry out tasks across Android apps
Photo: Santeri Viinamäki, CC BY-SA 4.0

Google says app makers can now let AI assistants like Gemini carry out tasks inside their Android apps, sometimes across several apps in one job. The idea is that you state a goal and the assistant handles the steps. The announcement came from Google's Android Developer Summit and is aimed at the people who build apps, but it shapes what your apps may do soon.

What changed

Google describes a new base layer called the Android intelligence system. It is meant to connect AI agents with apps while protecting users. It covers phones, wearables, XR glasses, TVs and cars, and offers help when you ask and help that starts on its own.

App makers connect through Android MCP, a tool framework based on a standard many AI tools already use. It lets apps work with approved AI agents right on the device. A separate remote option works across platforms.

Why on-device matters

Google recommends the on-device route when an app handles end-to-end encrypted data or sensitive information stored on the phone. It also fits tasks that need phone sensors or must run locally.

A newer small model, Gemini Nano 4, can now call functions, which lets apps build multi-step helpers that run fully on the phone. Google also offers a hybrid setup that uses cloud models when needed and the phone's model when available.

Gemini and Search

Google also introduced the Connected Apps Platform. It lets apps reach people through Gemini and Search, and Google says users are moving from asking questions to taking action. Developers can test locally today. A preview waitlist exists for production, but space is limited to selected apps in supported categories. Broader access is planned later.

  • Gemini can complete multi-step jobs, sometimes across apps.
  • Sensitive tasks can stay on the device.
  • Early access is limited to a small group of apps.

What is not known yet

The material does not say which apps will support this, when you will see it, or what permission screens will look like. It also does not explain how Google picks apps for the preview. Those details will decide how much control users actually get.

What to do

When an app or assistant asks for access to act for you, read what it wants. Booking or ordering on your behalf means giving an assistant more reach than a simple search does. Business owners with an app can look into the waitlist and local testing.

The takeaway

AI assistants may soon book and order through your apps, so check what access each app and assistant asks for.

More articles

You should really read these.

Next step

Let’s talk