Google’s experimental “bonobo” feature could let Gemini tap, swipe, and complete tasks inside apps—if users trust it.
Google is quietly preparing to give Android phones a far more active role in everyday digital chores. An experimental capability called Gemini screen automation, spotted in the Google app 17.4 beta, suggests Google’s AI will soon interact directly with apps—tapping buttons and swiping screens on users’ behalf.
Gemini’s Biggest Shift Yet: From Answers to Actions
If enabled, Gemini wouldn’t just suggest steps—it would do the work itself. Code strings reference a Labs feature dubbed “Get tasks done with Gemini,” internally codenamed bonobo, designed to control on-screen elements.
What does that mean in practice? You could ask your phone to order dinner or book a ride, without ever opening the app. It’s the difference between a GPS giving directions and a self-driving car taking the wheel—would you hand it over?
Limited Apps, Heavy Guardrails
Early indications show support will be limited to “certain apps” at launch. Think food delivery, ride-hailing, and core Google services—apps with high usage and predictable interfaces.
- This constraint likely reflects how often third-party apps change layouts, which can break automation.
- The feature appears to rely on Android 16 QPR3, which lays the system-level groundwork for deeper AI control.
Google hasn’t confirmed a release date or a compatibility list, and for now, the feature remains buried behind experimental flags.
Safety Warnings and Privacy Trade-Offs
Google’s own strings carry blunt warnings. “Gemini can make mistakes,” the company says, and users remain responsible for any action the AI takes. You’ll be able to interrupt the agent at any time and take over manually—but would you catch an error in time?
The company also urges users not to:
- Enter payment details or sensitive information into Gemini chats
- Rely on screen automation in emergencies
Privacy is another flashpoint. If the optional “Keep Activity” setting is enabled, screenshots captured during automation may be reviewed by trained staff to improve Google services. That transparency is explicit—but the trade-off is real.
Part of Google’s Broader Gemini Bet
Screen automation fits neatly into Google’s accelerating Gemini rollout. In recent weeks:
- Google Maps expanded Gemini support from driving to hands-free walking and cycling navigation
- Chrome’s Auto Browse began automatically filling forms and handling repetitive web tasks
The pattern is clear. Google is betting users will accept deeper automation if they retain control, visibility, and an opt-out switch. The open question: where is the line between helpful agent and overreaching assistant?
When Gemini screen automation escapes beta strings and starts acting on real phones, that answer will come fast.
TL;DR
Google is testing Gemini screen automation, an experimental feature that lets AI tap and swipe through Android apps to complete tasks. Spotted in the Google app 17.4 beta, it promises hands-free ordering and bookings, but raises clear safety and privacy concerns.
AI summary
- Signals Google’s push toward action-based AI agents
- Gemini may soon perform tasks directly inside Android apps
- Feature is experimental, limited to select services
- Strong warnings around errors, payments, and privacy
- Relies on Android 16 QPR3 foundations
