Gemini can now automate some multi-step tasks on Android

On Wednesday, Google announced a batch of new Gemini-powered updates coming to Android. The headline change is a new automation feature that lets Gemini carry out multi-step actions — like ordering an Uber or placing a food delivery — along with other updates rolling out today, including broader scam detection for phone calls and improvements to Circle to Search that make it possible to identify everything shown on your phone’s screen.

Google says these automations are meant to let people hand off parts of their to-do list to Gemini. In reality, though, the range of actions Gemini can handle is still fairly constrained at launch.

The company said the new capability is in beta and will initially work with a limited set of apps in the food, grocery, and rideshare categories. For now, it will only be available in the United States and South Korea.

In the U.S. and Korea, the supported apps include DoorDash, Grubhub, Instacart, Lyft, McDonald’s, Starbucks, Uber, and Uber Eats. In South Korea, Kaemin and Kakao T will also be supported.

At launch, Google says the feature will be restricted to the Gemini app on specific devices, including the Pixel 10 and Pixel 10 Pro, and the Samsung Galaxy S26 lineup.

Because AI-driven automations can potentially make mistakes, Google is building in safeguards. The automations won’t start without a clear command from the device owner. While they’re running, users can watch the steps as they happen and stop the process if the system encounters an error or appears stuck. Google also said the automations run within a secure virtual window on the phone, which can access only a limited set of apps, not the broader data stored on the device.

The update fits into a wider industry push to use AI to automate more personal and everyday tasks. ChatGPT, for example, allows users to set up tasks that run at specific times or on schedules, and also offers an agent that can perform various computer-based actions such as working with calendars, generating slideshows, or running code. Anthropic’s Cowork brings Claude’s capabilities to non-coding workflows, allowing non-developers to automate routine file and task management. An AI tool called OpenClaw recently went viral for handling everyday tasks, such as sending emails, managing calendars, checking flight status, and more.

Google also announced an expansion of its Scam Detection feature for phone calls. Call protection is now available on Samsung Galaxy S26 series devices in the U.S. (It already exists on Pixel phones in the U.S., Australia, Canada, India, Ireland, and the U.K.) Google is also using its on-device Gemini model to spot scam text messages in the U.S., Canada, and the U.K. on Pixel 10 series phones, and it says this will soon come to Galaxy S26 devices as well.

Another update involves Circle to Search, Google’s gesture-based search feature that lets users circle, scribble, or tap to start a search. Google says Circle to Search can now recognise and search for everything visible on the screen—not just a single object. That means, for example, you can search for every clothing item and accessory in an outfit you like, or dig deeper into a group of items and the broader topic they represent.

Google has been pushing Gemini updates into the Android ecosystem at a steady pace through operating system releases and Pixel-focused updates such as Pixel Drops. Meanwhile, Apple has been struggling to ship a more complete suite of AI features, including an AI-powered Siri, a rollout that was recently delayed again to later in the year.