Gemini can now book Uber rides and build DoorDash orders

Gemini’s new trick feels familiar in a way that’s both comforting and a little suspicious. A decade ago, Siri was supposed to call you an Uber, and Google Assistant was supposed to “order my usual.” Most of the time they just launched the app and left you staring at the same screens, with the small sting of “why did I bother.”

This time, Google is trying to finally cross the gap between asking and doing. On the Galaxy S26 launched recently, Gemini can take a multi-step request and work through it inside the app, and actually call an Uber for you. You long-press the side button, say what you want, and the automation runs while your phone stays usable. Google is also upgrading Circle to Search on the Galaxy S26 with multi-object recognition, so a single gesture can identify an entire outfit or scene at once, not just one item.

The way the mutli-step request runs suggests Google knows exactly where trust falls apart. Gemini opens the app in a secure virtual window that’s secure from the rest of your device. If you want to watch, you can. If you don’t, you keep texting, scrolling, doing whatever, and a live notification sits there like a progress meter for your errand. Tap it and you can see Gemini scroll, tap, and type in real time, with obvious controls that make it clear you can step in or stop it.

That same theme shows up again at the finish line. Gemini can get you to the booking stage, build the cart, line up the reorder, then it slows down right before the irreversible moment. You still confirm the fare. You still hit the order button. You even have to grant permission before the automation appears in the first place. Think of it as the assistant doing the boring setup, then handing you the final click which matters.

Sameer Samat at Google called these chores “digital laundry,” which lands because it’s exactly the kind of work you always mean to do and never enjoy doing. Rides, dinner orders, grocery carts aren’t exciting. They’re just repetitive enough that you might resent them, and just important enough that you can’t ignore them. 

You can still feel the early-preview edges, as It’s limited to a small set of food, grocery, and rideshare apps for now. It starts in the US and South Korea. And sounds like you can only run one automation at a time. One report says what Gemini sees is processed in the cloud, which is going to be the first fight the moment this gets widely used.

If this works, the bigger shift is not ordering tacos faster. It’s that the assistant starts to become the interface you actually use, and apps begin to feel less like destinations and more like utilities. The real test is whether Gemini stays steady when app layouts change, when restaurants run out of items, and when the clean path breaks in the middle. If it can handle those moments without turning into a question machine, this is the first mobile “agent” feature in years that might stick.