Google has started testing a feature that lets its Gemini assistant place phone calls to businesses on a user’s behalf, TechCrunch reports. The feature, called Call for Me, initially rolls out to Pixel 11 owners in the United States who pay for a Gemini subscription, and it requires the beta version of Google’s Phone app for Android.
The experiment puts Google in direct competition with Meta, whose Muse and Instinct agents can already make calls for users. Voice agents that handle routine calls, booking a table, checking store hours, asking about stock, are becoming the first consumer-facing form of agentic AI, and both companies want to own the default experience.
How it works and who gets it
The rollout is deliberately narrow. Pixel 11 owners only, United States only, paid Gemini subscription required, and the beta Phone app must be installed. Google is framing the whole thing as an experiment rather than a launch, which leaves room to change pricing, availability or behavior before a wider release.
That gating tells you something about how Google sees the risk. Making calls on someone’s behalf involves a business interacting with a system that may not disclose itself as an assistant, and the outcomes, a booked appointment, a wrong answer, a confused shop owner, are harder to roll back than a bad chatbot reply. Starting with a small, paying, self-selected group limits the blast radius while the company collects real-world data.
For users, the pitch is straightforward. Instead of calling three plumbers to compare availability, you ask Gemini, and it does the calling and reports back. The value depends almost entirely on how well the agent handles the messy parts of real phone conversations: accents, background noise, unexpected questions, and the moment when a business asks something the caller’s owner would need to answer personally.
The competitive picture
Meta’s Muse and Instinct agents are already making calls for users, which means Google is following rather than leading here. That is a familiar position for the company. It shipped Gemini’s image generation after competitors, entered the assistant market after Alexa and Siri, and still ended up with the largest distribution through Android.
The difference this time is that distribution alone may not decide the race. A calling agent lives or dies on call quality, and call quality is a data problem. The company whose agents make more calls gets more transcripts, more failure cases, and more chances to fix them. Meta moving first gives it a head start on that loop. Google’s counterweight is that Pixel owners are a small slice of the Android base, and a successful feature can scale to hundreds of millions of Samsung and Xiaomi phones in a way Meta’s agents cannot.
There is also a hardware angle. Pixel 11 exists partly to showcase exactly this kind of on-device AI, and Google has used flagship features before to define what the Android platform does next. If Call for Me works well, expect it in the Phone app’s stable release within a couple of cycles, at which point every Android manufacturer has to decide whether to keep it, reskin it, or build their own version on Gemini’s API.
What businesses will think
The feature also creates a new counterparty: every small business that suddenly receives calls from AI assistants. Restaurants, salons and repair shops have dealt with automated booking systems for years, but those were inbound systems the business chose to install. Call for Me is outbound and unsolicited.
Some businesses will welcome it as extra bookings without staff time. Others will find that AI calls waste the same minutes as spam. The industry answer so far has been disclosure norms rather than rules, and Google has not detailed how Call for Me identifies itself to the person picking up. That detail, buried in a beta rollout, may matter more for adoption than any feature on the user’s side.
There is a monetization angle too. A Gemini subscription is the entry ticket, which fits Google’s broader push to convert assistant usage into recurring revenue. If Call for Me proves useful, it becomes another line item on the subscription pitch, the way Apple bundles services into iCloud plans. If it proves annoying or unreliable, it stays a Pixel beta footnote.
Where this leads
Calling businesses is the first everyday task where an AI agent can complete real work end to end. Browsing the web, comparing options and typing a message all have friction the agent can shortcut, but a completed phone call is a finished transaction in a way few other agent tasks are. That is why both Google and Meta started there, and why the results of this experiment will be watched beyond the Pixel community.
The open questions are practical. Does the agent hand off cleanly when a call needs a human decision? Does it book the wrong date and leave the user to untangle it? How often do businesses hang up on it? Google will get answers from its beta group before any wider rollout, and Meta’s earlier start means the industry already has some baseline for what works.
Regulators may weigh in eventually too. Voice agents calling on a person’s behalf sit close to the line between assistance and impersonation, and several US states have rules about automated calls that were written before assistants could hold a real conversation. Any mass-market launch would likely need a clear disclosure standard, not just a settings toggle.
For now, the feature is a test with a small door. Pixel 11, US, paid subscription, beta app. But the direction is set. The next version of the phone assistant is not a better voice interface for your own calls. It is a caller that makes them so you do not have to.