Google unveiled a series of new Gemini features for Android on February 25th. The highlight: Gemini can now complete multi-step tasks in real apps – without you having to approve each step individually.
What Gemini Can Do Now
Imagine: you long-press the power button and say “Order me the same thing as last time on DoorDash.” Gemini opens the app, navigates through the menus, puts the order together, and waits for your final confirmation. The same works with Uber, Grubhub, and grocery apps.
Another scenario: your phone recognizes that you have a flight in two hours. A button appears on the lock screen to directly call an Uber to the airport. Or you’re planning a barbecue – Gemini can go through the guest list and compile a matching shopping cart.
How It Works Technically
Google has Gemini execute apps in a virtual, sandboxed window on the device. This means: Gemini has no access to the rest of your phone – only to the specific app that’s currently needed. You can track progress via notifications, intervene at any time, or cancel the process.
That’s a smart approach. Instead of trusting blindly, you can watch Gemini work.
Where It’s Available
Initially only in the US and South Korea. And only on specific devices: Pixel 10, Pixel 10 Pro, and the Samsung Galaxy S26 series. More devices and regions are to follow – but there’s no timeline for that.
Perspective
Google’s move shows where mobile assistants are heading. Apple has offered little comparable with Siri so far. Anthropic is targeting the desktop workplace with Cowork. Google is aiming at the smartphone – the place where most everyday tasks actually happen.
The limitation to a few devices and regions also shows that Google is being cautious. Agents that independently operate apps and spend real money need to work reliably. Otherwise, that Uber to the airport quickly turns into a pizza at the wrong address.
Sources: Google AI Blog, TechCrunch, 9to5Google