Gemini could soon order food deliveries or call an Uber for you, with code in a beta app pointing to new abilities.
Using a screen automation system, it's thought that this is the next step towards an agentic AI system on your phone.
Gemini might soon be able to help with a range of tasks thanks to a system called "screen automation". Code referring to these action has been found in the Google app, leading to more understanding of the company's plans for the AI system on Android.
Google announced Gemini as the replacement for Google Assistant in 2023, with the AI-powered assistant making big strides in 2025 to bring smarter interactions to your devices. But for many, Gemini has really found its place on the smartphone, while others wait for its arrival on TV and in the car.
Google has always talked about the agentic powers that it wants AI to offer – that is the ability to do things on your behalf. It sees a future where you'll ask Gemini to create an itinerary for a weekend away and once you've approved the suggestions, you'll get the AI to book hotels and make reservations at restaurants.
It's these latter actions that are seen as "agentic", ie., the AI is acting on your behalf, beyond just surfacing information for you.
According to 9to5Google, there are strings of code in the Google app 17.4 beta that refer to "Get tasks done with Gemini". It appears to be codenamed "bonobo", with Google explaining that it can help with "placing orders or booking rides, using screen automation on certain apps".
Using screen automation suggests that this isn't going to happen in the background, it's likely that it's all going to happen in front of your eyes, meaning you can intervene if things aren't heading in the right way.
How could an agentic Gemini help you?
While there's no working example, it's easy to imagine that asking Gemini to book you an Uber home would open the Uber app, fire up the destination, select a ride and tell you when it's arriving.
Equally, asking Gemini to order your usual from a delivery service should be fairly straightforward, as long as Gemini knows your preferences.
The source identifies a couple of additional pieces of information lurking in the app, including the declaration that "Gemini can make mistakes" and that "you're responsible for what it does on your behalf".
This latter phrase is a great disclaimer, so that if Gemini orders a stack of pizzas, you can't then turn around and say that you didn't want them and the AI is to blame. I think we all know that's going to happen a lot in this agentic future.
It's said that you'll be able to intervene and stop the AI at any point, but that does sort of beg the question of why you'd automate this process in the first place? Is it so arduous to open an app (which Gemini will already do for you) and make a few taps?
It's said that "screenshots are reviewed by trained reviewers and used to improve Google services if Keep Activity is on" which will raise privacy questions with some users, while it also says "don’t enter login or payment information into Gemini chats", suggesting that it can't be trusted with sensitive information.
Like all things AI, the big question comes with whether this is a good thing or not. We know that AI is gaining skills all the time, with the ultimate aim to carrying out tasks on your behalf in the future.
But if it can't be trusted to get it right and there's a chance that personal information might not be secure, is this ever going to get off the ground?
