Samsung’s Galaxy S26 Gets Gemini ‘Screen Automation,’ Letting AI Order Coffee and Book Rides

On a recent afternoon in San Francisco, a Galaxy S26 Ultra owner could order a round of customized Starbucks drinks and hail a ride without touching the screen more than once.

“Order a Medicine Ball from the nearest Starbucks for pickup,” the user says, invoking Google’s Gemini assistant. A small window pops up over the display. The phone opens DoorDash, searches for Starbucks, picks the drink — officially called Citrus Honey Mint Tea — adjusts the size and sweetener, and lines up the payment. All that is left is a single tap to confirm.

That scenario, documented in early hands-on tests by Android-focused reviewers, is no longer a concept demo. It is the centerpiece of a new screen automation feature that began rolling out this month on Samsung’s Galaxy S26 lineup, marking one of the first mainstream deployments of a general-purpose AI agent that can operate other apps on a user’s behalf.

From assistant to agent

Samsung unveiled the Galaxy S26, S26+ and S26 Ultra on Feb. 25 at its Galaxy Unpacked event in San Francisco and began selling the phones globally around March 11. Within days, Samsung and Google quietly turned on a new capability in Gemini, Google’s AI assistant, for S26 owners in the United States and South Korea: the ability to control apps and complete multi-step tasks by driving the phone’s on-screen interface.

The change moves smartphone AI from answering questions and opening apps into actually taking actions across them — ordering food, booking rides and groceries, and filling out forms — while raising fresh questions about privacy, liability and the growing power of platform owners to mediate everyday digital transactions.

Samsung calls the S26 its “third-generation Galaxy AI phone,” extending a strategy that began with the Galaxy S24 in 2024. At the February launch, TM Roh, president and head of Samsung’s Mobile eXperience business, described the company’s goal as shifting artificial intelligence from novelty to infrastructure.

“AI must become infrastructure that works for everyone, everywhere,” Roh said at the event, adding that Samsung wants to close “the gap between what AI promises and what people actually experience.”

How screen automation works

On the S26, that experience centers on an expanded role for Gemini. The assistant can be summoned with a long press of the power button, the familiar “Hey Google” hotword or the Gemini app. With screen automation turned on in settings, users can issue broad requests such as “Find the best coffee nearby and order it for pickup” or “Get me a ride home and something for dinner.”

Gemini responds by opening a virtual window that shows a live feed of the device’s screen. Within that frame, the assistant taps buttons, scrolls lists and enters text much as a human would.

At launch, supported apps in the United States include Uber, Lyft, DoorDash, Uber Eats, Grubhub and the Starbucks app, with grocery service Instacart listed as coming soon. In South Korea, a different set of local partners is expected, though full details have not been publicly listed.

While the AI is working, users see each step and can interrupt at any time with a prominent “Stop task” button or tap “Take control” to resume manual use. Before any purchase or booking is finalized, Gemini pauses on a confirmation screen and asks the user to approve the action.

Early testers have described the capability as a visible change in how phone assistants behave. One reviewer wrote that watching the system place a Starbucks order from an informal drink nickname was “pretty wild” and “one of the most capable agentic features introduced in a while.”

Limits and subscription tiers

The feature is limited in several ways. It currently works only on the Galaxy S26 series, requires Samsung’s One UI 8.5 software and at least the February 2026 security update, and is restricted to English-speaking users ages 18 and older in the United States and South Korea. It also relies on a personal Google account; work and school accounts are excluded.

Access is also tiered by subscription. People using Gemini for free can run only a handful of automated tasks each day, while customers paying for Google’s Gemini Ultra plan receive a significantly higher daily allowance. Google has not disclosed precise numbers for every market, but reviewers have reported limits around five automation requests per day for free users and up to 120 for paying subscribers.

Security claims—and unanswered privacy questions

Under the hood, Samsung and Google each provide part of the stack that enables the new behavior.

Samsung emphasizes device-level security as a foundation for letting an AI agent interact with sensitive apps. The company’s Personal Data Engine is designed to learn user preferences on the device rather than in the cloud. Knox Vault, a hardware-backed secure area, stores biometric data, passwords and encryption keys apart from the main operating system. A newer feature called Knox Enhanced Encrypted Protection (KEEP) isolates data within individual apps.

Although these systems were in place before screen automation, Samsung has presented them as essential to protecting logins and payment credentials when an assistant is driving the interface.

Google, for its part, has framed screen automation as an experimental or “Labs” capability within Gemini. The feature is disabled by default. To use it, S26 owners must open the Gemini app, go into settings and explicitly enable screen automation, choosing whether the assistant should ask for permission every time or be allowed to act more freely.

The company has not publicly detailed whether the assistant’s reading of on-screen content and decision-making happens entirely on the device or involves cloud processing. It also has not spelled out how long any logs or screenshots from automated sessions may be retained, or how they may be used to improve services.

Those gaps have drawn attention from privacy advocates and legal scholars focused on AI.

The new feature effectively gives Gemini visibility into whatever appears on the screen while it is active, from group chats and contact lists to maps and payment pages. If the AI is using that information to interpret instructions — for example, scanning a chat to see what friends want to eat — questions arise about consent from other people in the conversation and about how sensitive categories of data are handled.

The possibility of financial mistakes adds another layer of complexity. If Gemini misreads a message, selects the wrong restaurant or sends a car to the wrong address, it is not yet clear who would be responsible for the cost: the user, the platform operator, the underlying app or some combination.

Lawyers and regulators are also likely to scrutinize what happens when automated agents click through terms of service or consent dialogs, an increasingly common part of online transactions. Under consumer protection laws in the European Union and the United States, companies must ensure that key rights and obligations are clearly communicated. Whether a virtual assistant’s tap on “I agree” carries the same legal weight as a person’s informed consent is an open question.

A shift in platform power

The arrival of screen automation also has implications for competition among apps and platforms.

In earlier generations, assistants such as Google Assistant and Samsung’s Bixby typically relied on explicit integrations. A user who asked Assistant to “order an Uber” would trigger an application programming interface (API) call to Uber’s systems, or be handed off to the app with a prefilled destination. Bixby Routines, Samsung’s rule-based automation system, could open apps or adjust settings but not navigate arbitrary interfaces.

By contrast, Gemini’s new control layer is largely app-agnostic. It can, in principle, learn how to operate any supported app’s interface, turning companies like Uber, Lyft and DoorDash into interchangeable backends for the assistant. When a user simply asks for “the fastest ride to the airport,” the decision about which service to use, and on what terms, can shift toward the platform-level AI.

That intermediary role may give Google and Samsung greater influence over which services people use and how often they use them. It may also help the companies collect more comprehensive behavioral data across multiple apps, though both emphasize they are working to protect user privacy.

For now, screen automation remains limited in scope, device support and geography. But Google has already begun extending the feature beyond Samsung’s phones, adding it to its own Pixel 10 series through a March software update. If the rollout continues, the ability to ask a phone not just to find information but to carry out tasks inside other apps could quickly become a standard expectation for high-end devices.

Samsung’s Roh has described that vision as AI receding into the background, becoming as taken for granted as electricity or broadband. For Galaxy S26 owners in the first wave of countries, that future may already be starting to appear on the other side of a small virtual window floating above their screens.