Google’s Gemini AI takes control of Android tasks

Google has unveiled a major advancement in its Gemini artificial intelligence assistant that allows the software to autonomously handle multi-step operations on Android devices, enabling users to command the AI to book rides, order food and manage other everyday tasks without manual execution. The new capability positions Gemini not just as a conversational helper but as an active agent capable of executing instructions across multiple apps in sequence, a shift that could reshape the way people interact with mobile devices and intensify competition with other AI ecosystems.

At the core of this upgrade is Gemini’s ability to run supported third-party apps within a secure, virtual environment and complete processes that previously required users to switch between interfaces. For example, a user can ask Gemini to book a ride through services such as Uber or to arrange a meal delivery, and the AI will navigate the required applications, inputting details and following through to completion while allowing the user to monitor progress or intervene if necessary.

The rollout of these agentic features has begun on flagship Android devices including the latest Samsung Galaxy S26 series and Google’s own Pixel 10 phones, with the broader beta intended to expand to more hardware over time. At the announcement at Samsung’s Unpacked event in San Francisco, Samsung executive TM Roh described the integration as an “agentic AI phone” experience, emphasising the deeper role of artificial intelligence in simplifying routine tasks.

Underpinning this advancement is Google’s strategic push to make Android an “intelligent OS,” where artificial intelligence is woven into the operating system’s core functionality rather than remaining an add-on. The company’s Android development team has outlined a vision where AI agents can meaningfully reduce the friction involved in mobile workflows by interfacing with app ecosystems through frameworks such as AppFunctions, which establishes structured communication pathways between AI agents and installed applications. This development hints at a future where a wide range of capabilities — from calendar management to note organisation — could be automated by the assistant itself.

The automation feature is initially limited in scope. Support at launch focuses on a set of popular rideshare, food and grocery delivery applications, and the service is being offered in select markets including the United States and Korea. Google has emphasised that user consent is required before any automation begins, and that tasks run in a restricted environment that can only access the specific apps involved, safeguarding broader personal data on the device.

The move reflects a broader trend in the technology industry towards agentic artificial intelligence, where software increasingly moves beyond responding to queries towards completing full tasks on behalf of users. Competitors such as OpenAI and Apple are advancing their own versions of proactive AI agents; for instance, Apple has been expanding automation capabilities in its operating systems while OpenAI has promoted agent frameworks that can perform a range of activities based on high-level user instructions. Google’s introduction of task automation for Gemini further accelerates this shift, pushing AI deeper into everyday digital interactions.

Industry analysts note that this evolution could unlock substantial utility for users by reducing the cognitive load and repetitive actions associated with digital routines. Being able to automate a sequence of actions — such as planning a trip that involves ride-hailing, reservations and calendar entries — with a single instruction could mark a departure from traditional keyword-based assistants. Yet it also raises complex questions around user expectations and control, including how accurately an algorithm interprets nuanced requests and how developers ensure transparent operation without unintended consequences.

Privacy and security experts have flagged potential risks associated with granting AI agents broader access to personal devices. Research into task-executable assistants suggests that without careful implementation, such technologies could expose data or elevate privileges beyond what users anticipate, particularly when navigating between applications that hold sensitive information. Advocates for responsible AI underscore the importance of clear user consent, visible control mechanisms and robust safeguards as these systems evolve.