Google's Gemini assistant can now open apps, tap buttons, and complete real tasks on your phone - no hands required. The feature, called task automation, is rolling out first on Samsung's Galaxy S26 and Google's Pixel 10 devices.
The initial scope is narrow but practical: food delivery and rideshare apps. You tell Gemini you want dinner from your favorite Thai place or need a car to the airport, and it handles the rest inside a virtual window. It navigates the app, fills in details, and places the order or books the ride on your behalf.
This is different from Gemini just answering questions or summarizing text. Task automation means the AI is actually interacting with third-party apps the way you would - scrolling, selecting, confirming. Google and Samsung announced the capability a couple weeks ago, and early demos show it working across DoorDash-style delivery flows and Uber-style ride booking.
What's Actually Happening on Your Phone
Gemini runs the target app in a sandboxed virtual window, separate from your normal app usage. It reads the screen, identifies UI elements, and takes actions step by step. Think of it as a very patient friend who knows how to use every app on your phone and will do the boring parts for you.
The sandboxed approach matters for security - Gemini isn't getting free rein over your entire phone. It operates in a contained environment for the specific task you requested.
The Bigger Picture for Mobile AI
Google isn't alone here. Apple Intelligence has been slowly adding Siri capabilities, and Samsung's own Galaxy AI features have been expanding. But Gemini's approach of directly operating inside third-party apps is more ambitious than what competitors have shipped so far.
The food-and-rides starting point is smart. These are repetitive tasks with predictable flows - you pick a restaurant, choose items, confirm delivery address, pay. There's little ambiguity. Expanding to more complex apps (banking, travel booking, healthcare) will be the real test of whether this works reliably enough that people actually trust it.
For now, this is limited to two phone lines and two app categories. But if the execution holds up, it sets the template for how AI assistants evolve from answering questions to actually doing things.