San Francisco, 02:04 PST, February 4, 2026
- Code found in a Google app beta hints at Gemini’s “screen automation” feature, which can handle tasks like placing orders and booking rides within certain Android apps
- The warning text clarifies that users remain responsible for the agent’s actions and can interrupt it at any point during the task
- Privacy notices mention that if “Keep Activity” is turned on, trained reviewers might check screenshots
Google appears to be edging nearer to enabling its Gemini assistant to handle tasks within Android apps, like ordering items and booking rides. This development came to light after new code was spotted in a beta release of the Google app, according to tech outlet 9to5Google. (9to5Google)
That move would take Gemini past simple chat, turning it into an “agent” that can actually handle tasks for users. The timing’s crucial—businesses are racing to build AI assistants that do more than just respond; they’re meant to tackle real chores. And phones? They’re where most of those tasks pile up.
Google enabling Gemini to interact with third-party apps would give it more control over Android’s main interface. But that move also brings risks: costly errors, tricky fringe cases, and privacy concerns from letting an assistant access what’s displayed on screen.
Android Authority spotted strings in the Google app 17.4 beta pointing to a Labs feature named “Get tasks done with Gemini,” internally dubbed “bonobo,” which leverages “screen automation” within select apps. The warning is clear: “Gemini can make mistakes. You’re responsible for what it does on your behalf, so supervise it closely.” It also notes users can intervene and take control manually. (Android Authority)
Screen automation basically means controlling the UI directly: the assistant watches the display and interacts by clicking and typing just like a human user. It goes beyond neat app integrations and can get a bit messier.
Android Police revealed the beta text includes a privacy warning: “When Gemini interacts with an app, screenshots are reviewed by trained reviewers and used to improve Google services if Keep Activity is on.” Users are urged not to input login or payment details during Gemini chats and to steer clear of screen automation for emergencies or sensitive actions. (Android Police)
SamMobile reported the basic idea is straightforward: ask Gemini to order an Uber or get food delivered, and it handles the process within supported apps on a Galaxy device. The site noted that initial support could be limited to just a few apps, though that may grow over time. (SamMobile)
This feature seems linked to Android 16 QPR3, which stands for “Quarterly Platform Release,” Google’s routine update cycle following a major Android launch. So far, Google hasn’t revealed which apps will support it or when this Labs experiment might roll out to all users.
Another report from Hindustan Times highlights that the same beta build mentions a “Likeness” feature, codenamed “wasabi,” tied to 3D avatars in Google Meet on Android XR. The text implies users might control their likeness via prompts, adding that the likeness “can only be used” by the individual user. (Hindustan Times)
The push enters a packed market. Apple is framing Siri improvements under its Apple Intelligence initiative, while Samsung and others rush to load AI features directly onto devices. Google’s Gemini has to feel both natural and practical, or competitors will set the standard for what an “AI phone” actually means.
The risk shifts once an assistant moves from drafting text to hitting “Buy.” Reliability becomes the main question: a wrong tap, misinterpreted screen, or a confusing checkout process could lead to incorrect orders, accidental charges, or accidental data exposure. Since this comes from early-stage code, it might never launch—or arrive only in a limited version.
