Google tests Gemini agent that can call businesses and handle tasks for users

Google is testing a new Gemini feature that can call businesses on a user’s behalf, handle automated phone systems, wait on hold, and complete selected tasks such as reservations or appointment changes. Called “Call for Me,” the experiment initially targets Pixel 11 owners in the United States who pay for a Gemini subscription.

The feature marks a shift from AI-assisted phone navigation toward delegated action. Instead of merely helping a user find a business number or move through an interactive voice response system, Gemini can conduct the conversation and share personal information that the user has approved. For now, however, access is limited and the system remains experimental.

A narrow US experiment

According to reporting by TechCrunch AI and The Decoder, Call for Me requires both a paid Gemini subscription and the beta version of Google’s Phone app for Android. Google is starting with Pixel 11 devices in the US rather than making the capability broadly available across Android phones or markets.

Google said the limited launch reflects the difficulty of handling real-world conversations. Business calls can involve ambiguous requests, unexpected questions, poor audio, transfers, and policies that are not clearly stated in advance. The company is therefore treating the feature as a test while it evaluates how reliably Gemini performs outside controlled demonstrations.

Call for Me also uses the user’s own phone number to place calls directly from the phone. That detail could make the interaction appear more familiar to businesses, but it also makes consent, disclosure, and responsibility more important when an AI is speaking to staff or making commitments on a customer’s behalf.

From lookup to transaction

Google says Gemini can use approved personal information to perform a broader range of actions than its earlier calling tools. Potential examples include asking whether a product is in stock, placing an item on hold, booking a restaurant, or moving an existing appointment to another date.

The agent is designed to introduce itself, navigate phone menus, remain on hold, and continue the conversation with a business employee. Users can monitor the exchange through a live transcript and take over at any point, according to the company’s description reported by both publications.

That combination of delegation and intervention is important. A fully autonomous call could be efficient but difficult to trust if the system misunderstood a date, price, name, or cancellation policy. A live transcript gives users a way to detect errors, while the takeover option provides an escape route when the conversation becomes too complex.

Call for Me builds on Google’s longer-running work in this area. Google Duplex demonstrated automated restaurant and salon calls several years ago, while Pixel users have also received tools including Hold for Me, Talk to a Live Rep, and Direct My Call. Google’s more recent Ask for Me feature focused on calling businesses to gather information such as prices or available services.

Evidence and limits of the claim

The available reporting establishes that Google is testing Call for Me and describes the company’s intended capabilities. It does not provide independent evidence of completion rates, error rates, call duration, business acceptance, or user adoption. Claims about what the feature can handle should therefore be treated as Google’s product description, not as verified performance results.

The small rollout also means there is not yet enough evidence to judge whether Gemini can reliably manage the wide variety of conversations involved in retail, hospitality, healthcare, or service businesses. A task such as checking inventory may be relatively straightforward, while rescheduling an appointment could require identity verification, policy interpretation, or payment-related decisions.

The competitive context is moving quickly. TechCrunch AI reported that Meta’s Muse and Instinct can also make calls for users, while The Decoder connected Google’s test to earlier AI calling efforts. These comparisons show that AI agents are expanding from text and app interfaces into voice interactions, but the sources do not establish which system performs best.

Why the test matters for builders and enterprises

For product teams, Call for Me illustrates a more demanding form of agent design. The system must combine speech recognition, dialogue management, phone-system navigation, user permissions, personal data handling, and recovery when a call goes off script. The difficult engineering problem is not simply generating a natural-sounding voice; it is maintaining reliable control over a real transaction.

Enterprises will also need to consider how AI-originated calls are identified and handled. If multiple agents begin contacting businesses, customer-service teams may face more automated conversations, repeated requests, and uncertainty about whether a caller is authorized to make a purchase or change an appointment. Businesses may respond by adding stronger verification or by restricting automated calls.

For consumers, the appeal is clear: Gemini could remove the time spent listening to menus and waiting on hold. But delegation creates risks around incorrect information, accidental commitments, privacy, and social friction. The ability to approve information in advance, watch the call, and intervene may become as important as the agent’s ability to finish the task.

What to watch next

The first signal will be whether Google expands Call for Me beyond Pixel 11 owners, paid Gemini subscribers, and the beta Phone app. Broader device and geographic availability would indicate that the company is moving beyond an early experiment.

More important will be evidence about reliability: how often calls reach the correct department, how frequently users need to take over, whether businesses accept AI callers, and how Google handles failed or disputed transactions. Product documentation should also clarify which personal details can be shared, how consent is recorded, and whether calls are clearly identified as AI-assisted.

Competitors’ rollout strategies will provide another test. Meta’s Muse and Instinct may push voice delegation into more consumer products, while Google’s advantage could come from integrating the capability directly into Android’s Phone app and Pixel hardware.

Creati.ai perspective

Call for Me is a meaningful test of whether consumer AI agents can move from answering questions to completing ordinary offline tasks. Google has the distribution, phone integration, and years of prior experiments to make the use case plausible, but the narrow launch shows that dependable real-world conversation remains unresolved.

The feature’s success should be judged less by how natural the voice sounds than by whether users can safely delegate a task, understand what happened, and recover when Gemini makes a mistake. Until Google publishes broader usage and reliability evidence, Call for Me is best viewed as an important product experiment rather than proof that phone-based agents are ready for routine autonomy.