Google’s Gemini AI can now handle multi-step tasks on Android, making it more like an on-device agent that can get things done across apps rather than just answering prompts in a chat window. Google says the feature is currently in early preview on a limited set of flagship phones and in a small set of markets.
For now, multi-step tasking focuses on everyday to-do list items; users can ask Gemini to order their favorite takeout, schedule a grocery pickup, or book a cab. The assistant then navigates supported apps on their behalf and stops at a confirmation screen.
Google built the feature on an “intelligent OS” approach for Android, where users can long-press the power button to open Gemini and provide natural-language instructions. Gemini follows these instructions in virtual windows across supported apps. It taps into existing app data, such as saved addresses and order histories, to speed up the process, but it saves actions like payments for manual approval.
The early rollout targets phones like the latest Pixel and Samsung Galaxy S devices in the US and South Korea. It only works with a limited set of partner apps, though we can expect that list to expand in the future.
Google positions this as part of a broader push toward “agentic” experiences, in which AI can execute tasks on a user’s behalf. On one side, you have the Android version of Gemini starting to handle real, multi-step errands inside apps; on the other, Siri still struggles with some of the most basic queries. Perhaps Apple’s recent deal with Google to produce a Gemini-powered version of Siri will help.
