Gemini begins making phone calls on behalf of users, AI agents breach traditional telecom networks

Google has begun testing a new feature that lets its Gemini model place full phone calls with companies and stores on users' behalf, called “Call for Me.” The trial, in its first phase, is limited to owners of Pixel 11 phones who are subscribed to paid Gemini plans in the United States and who use the experimental Android phone app, a move that coincides with rival smart agents such as Meta’s “Muze” and the “Instinct” agent entering the arena of voice calls on behalf of their owners.

This feature goes beyond the earlier voice assistants the company has offered for years, which were limited in capability. In this update,The smart agent has the authority to share personal data that the user has pre-approved in order to complete complex tasks,Examples include calling a store to verify that an item is in stock, reserving a table at a restaurant, moving a previously scheduled appointment to another date, or requesting a product reservation at a point of sale. The system places the call directly from the user’s personal phone number, introducing itself to the other party, navigating automated menus and their sub-options, waiting on the line until a customer-service representative answers, and then handling the entire conversation.

To ensure control and prevent deviation during the call, the app lets the phone owner monitor the exchange in real time via a live text transcript that appears on the screen, while the user retains the ability to intervene and take over the call at any moment. Google bases this step on a long-running experiment that began when its voice assistant demonstrated the ability to book a salon appointment using natural-sounding human speech patterns, followed by the launch of the “Ask for Me” feature last year for price inquiries, as well as earlier Pixel capabilities such as “Hold for Me” and “Talk to a Live Rep” for waiting on behalf of the user, and the “Direct My Call” function to simplify navigation of interactive response systems.

Google justified the current restriction of the feature to a limited trial by noting that real-world conversations are complex and involve subtle variations, which require time to test the system’s accuracy before a broader rollout. Nevertheless, the significance of this development goes beyond smartphones; it moves AI models from text-chat rooms into the infrastructure of traditional voice-call networks, turning the model into an executive agent that operates through the user’s actual phone line.

For technology and operations leaders in the Gulf, Egypt and the wider region, this shift carries early signals of a reshaping of customer-service systems and call centres. When inbound calls originate from real customer phone numbers but are spoken by a smart agent that conducts the conversation, executes tasks and bypasses queues in seconds, support centres will find themselves forced to handle non-human callers demanding precise answers without hesitation. This reality compels organisations, especially in retail, banking and hospitality, to accelerate the upgrade of interactive voice-response systems or connect them to direct APIs that enable automatic understanding between client agents and corporate systems without draining human staff time.

The trend also adds extra burdens to information-security policies and local identity-verification procedures, as authorising an agent to share personal data during a phone call raises legal and regulatory questions about the reliability of the consent and the verification of the synthetic speaker’s identity. The encroachment of AI agents into daily call management means that business models that have relied for decades on keeping the human customer on hold will be the first to erode in the forthcoming digital-communication landscape.