Google’s latest experiment pushes Gemini from answering questions into doing a familiar, frustrating task: making the call. But the company is rolling it out cautiously, limiting access while it tests whether an agent can cope with the untidiness of real conversations.

The feature, called Call for Me, follows years of Google work on AI-assisted calling. Earlier tools could screen calls, map automated phone menus or wait on hold; last year’s Ask for Me allowed Gemini to contact businesses for information such as prices and services. The new version is designed to go further, letting the assistant act on a user’s behalf in a wider range of exchanges.¹

Initially, the experiment is available only to U.S. Pixel 11 owners with a Gemini subscription and the beta version of Google’s Phone app. Gemini will place calls using the customer’s own number, introduce itself to the business, work through automated menus and wait on hold. Users can watch a live transcript and step in at any point.¹

Google says the agent can tackle jobs that often consume time but require little creativity: checking whether a store has an item in stock, booking a restaurant table, rescheduling an appointment or asking a business to hold merchandise. Crucially, it can also relay personal information that the user has approved, expanding what it can accomplish during the call.¹

That promise comes with a conspicuously tight leash. Google says it is beginning at small scale because “real-world conversations are nuanced,” a recognition that phone calls can turn on unclear requests, unexpected questions and the judgment of the person on the other end.¹ Call for Me is therefore less a declaration that AI has mastered the telephone than a controlled test of whether it can reliably handle the waiting, menu-diving and routine negotiation humans would rather avoid.