Google is apparently pushing Gemini to the point where it will no longer just answer, but will actually perform tasks directly on your Android phone. In a new beta version of the Google app, strings of code appear that describe the "Get tasks done with Gemini" – internally under the code name bonobo. In practice, it is screen automation, i.e. the possibility for Gemini "saw" the screen and completed the task step by step in supported applications.
The description from the beta version makes it clear – Gemini should be able to handle place orders or maybe book a ride and similar routine things. And that's just based on what you have on the display. The important detail is that it won't apply universally everywhere, but only in "certain" apps “, i.e. in selected applications. There have been earlier indications that Android 16QPR3 is paving the way for this progress, and now Google has actually come to make Gemini virtual assistant.
The key is that this is not a "magic wand" that runs without control. Google directly warns that Gemini can make mistakes and that you are responsible for what he does on your behalf. So there is a regime in place where you can hire an agent stop at any time and complete the task manually.
Gallery
The biggest question logically concerns privacy. Google states that when Gemini interacts with the application, screenshots can be reviewed by trained evaluators and used to improve the service – if you have activity saving turned on. At the same time, it adds a very practical warning:
- do not enter into chatlogin or payment details
- do not use automation for sensitive situations or emergency scenarios
Change in the Ge web applicationmini
The change in the Ge web application also fits into the same mosaic.mini. A new section appears in the Tools menu Labs, which separates experimental features from regular ones. And next to that comes a toggle "Personalize" chat when helpful", which decides whether Gemini will reach to Connected appsThe important thing is that this only applies to the current conversation, so you can choose when you want more context and when you want a clean, unconnected answer.
Overall, it looks like Google is moving towards a model where Gemini to act as an AI agent, which not only advises, but also acts. And at the same time, it establishes clear boundaries so that the user can better understand, what is an experiment and where it is a function ready for regular use.
