Google tests Gemini making calls for Pixel users

Google is experimenting with a feature that would let its Gemini assistant place phone calls on behalf of Pixel users. According to the company, the capability will initially reach owners of the Pixel 11 in the United States who subscribe to a paid Gemini plan. The idea is to turn the AI into a task runner for routine errands: booking appointments, checking business hours, or making reservations. Instead of waiting on hold or navigating automated menus, users could ask Gemini to handle the call and report back. The feature fits a broader industry push toward agentic AI, where assistants do more than answer questions and instead take actions across services. For consumers, the appeal is obvious: saving time on mundane chores. But live phone calls introduce complications that chat-based assistants do not face. Businesses may not know they are speaking with an AI unless it identifies itself. People on the other end could be recorded or transcribed, raising consent and privacy issues. There are also questions about accuracy: an AI might misstate a request, agree to the wrong time, or fail to understand an accent or a noisy line. Google will need to balance helpfulness with disclosure. If Gemini sounds human, callers might be misled. If it announces itself too rigidly, some businesses might hang up or refuse to engage. The company will also have to explain what data is collected from calls, how long it is stored, and whether users can review or delete it. Regulatory attention around AI deception and automated decision-making is growing, so early design choices could set expectations for rivals. For now, the test appears limited to a specific device and subscription tier, which gives Google room to learn before a wider launch. The outcome will depend on whether the assistant can complete tasks reliably and whether businesses and regulators accept AI voices in real-time conversations. If it works, agentic calling could become a standard smartphone feature; if it stumbles, it may become a case study in why some automation needs clearer boundaries.