What Call for Me does
The new version expands what Google’s calling agents can do. Gemini may share personal information approved by the user during the conversation, giving it more room to complete tasks such as:
Gemini will identify itself to company staff, work through an automated phone menu and wait on hold. When a person answers, it can continue the conversation. The user follows the call through a live transcript and can take control at any point.
Calls are placed from the user’s personal number and directly from the user’s phone. That setup keeps the interaction visible rather than turning it into a fully detached background service.
A deliberately small launch
Google is limiting the experiment to Pixel 11 owners in the US who subscribe to Gemini. Participation also requires the beta version of Google Phone for Android.
The company says real-world conversations depend on too many nuances to expand the feature immediately. Google wants to refine Call for Me before making it more widely available.
That explanation is significant: the hard part is not dialing or transcribing. It is handling the unpredictable handoff between an automated system, a business employee and information the user has authorized Gemini to share.
Google has been rehearsing this for years
Call for Me extends a line of phone experiments that Google has been developing over several years.
At an earlier I/O developer conference, Google showed Assistant calling a salon to book an appointment. To make the exchange sound more natural, the system inserted short pauses and fillers such as “um” and “uh.”
Last year, Google introduced Ask for Me, which allowed Gemini to call businesses on the user’s behalf to find out prices or available services.
Pixel owners also have access to a set of narrower calling tools:
My read is that Call for Me is less a new category than Google’s attempt to connect these separate utilities into one agent. The more consequential change is the permission to use approved personal information while negotiating with a business.
The announcement is quiet about what kinds of information users can approve and how those permissions are managed. That boundary will matter more than the ability to wait on hold: Google is testing whether an agent can act in a user’s name without making the user surrender the call.
Daily AI news
Every day we pick what actually matters in AI and explain it plainly — no hype, no filler. Subscribe if you want to follow where the industry is going.
Only what matters — every day
Follow on X