Your phone can now make its own calls
Google has a new idea. It is for your phone. Your phone should make its own calls. The company announced on 24 September a new feature that allows its artificial intelligence, called Gemini, to place phone calls to local businesses entirely on your behalf. It handles the whole process. You do not speak. The AI does. This is not science fiction. It is an experiment. It is starting now.
The system is designed for convenience. A user does not need to start the call themselves. You will not even hear a dial tone. Instead, you give Gemini an instruction through its app. You might type, ‘Book a table for two at that new Italian place for 8pm Saturday’. Or perhaps, ‘Check if the bike shop has any 24 inch inner tubes’. Gemini takes that instruction. It finds the number. It makes the call. The phone in your hand remains silent while its digital assistant gets to work, a ghost in your own machine placing orders and asking questions for you.
Gemini handles the tedious parts. All of them. It introduces itself to the business. It navigates the automated phone menus that guard the gates of so many customer service lines, pressing one for this and two for that. Google claims the AI will patiently wait on hold. This is the feature’s central promise. It frees the user from the purgatory of listening to crackly hold music while waiting for an employee to answer. The AI will absorb the wait so you do not have to, a small but significant delegation of modern digital life’s many tiny frustrations.
You are not completely detached from the process. You are a silent supervisor. The company says the user remains in control at all times. This is achieved through a live transcript that appears on the phone’s screen. You can watch the conversation unfold in real time, reading the words spoken by both Gemini and the human on the other end of the line. If something goes wrong, or if you simply want to take over, you can do so at any moment. It is your call. You are just outsourcing the talking.
The tasks are deliberately ordinary. They are small and local. You could ask Gemini to make a dinner reservation, reschedule a dental appointment, or check if a shop has a specific product in stock before you make the journey. These are simple, goal oriented conversations. They have a clear beginning, middle, and end, making them ideal training grounds for an AI still learning how to interact with the messy, unpredictable world of human commerce and conversation.
This is not for everybody. Not yet. Google is framing this as an ‘early experiment’. Access is restricted. For now, only people in the United States with the new Pixel 11 phone will be able to use it. There is another condition. They must also be paying for a Gemini subscription. It is a tiny test group for a big idea.
This sounds familiar for a reason
This idea is not new. Google has tried it before. The company is revisiting a very old, very controversial ghost from its recent past, a project that generated both wonder and a deep sense of unease. The year was 2018.
Sundar Pichai, Google’s chief executive, stood on stage at the Shoreline Amphitheatre in California during the company’s annual developer conference. He played a recording. It was a phone call. An AI assistant, then called Duplex, booked a haircut appointment with a real salon employee who was completely unaware she was speaking to a machine. The AI sounded flawlessly human. It peppered its speech with conversational tics, the little ‘umms’ and ‘ahs’ we all use to signal thought and understanding, a detail that made the demonstration both brilliant and deeply unsettling. A second call followed, this time to a restaurant. The AI handled the interaction perfectly. The audience cheered. It was a stunning technical achievement. It was also terrifying.
The backlash was immediate. It was fierce. Ethicists, journalists, and ordinary people recoiled from the deception at the heart of the demonstration. Was it right for a machine to fool a person into believing it was also a person? The salon receptionist had not given her consent to be a research subject for Google’s latest experiment. She had not agreed to train an algorithm. She just answered the phone. The fundamental contract of a phone call, an interaction between two conscious individuals, had been broken without warning. The perfect, seamless imitation was not a triumph of engineering. For many, it was a breach of trust.
Google was forced to retreat. It was a significant and public climbdown. The company issued a statement, promising that any Duplex call would henceforth begin with a clear disclosure, informing the recipient they were speaking to an automated system. The grand vision of an AI indistinguishable from a human was abandoned. The project went quiet. Duplex technology was eventually scaled back, its power relegated to less glamorous tasks like automatically updating business opening hours on Google Maps. The ambition, however, never disappeared. This new Gemini feature is Duplex resurrected. It is the same core idea, powered by more advanced models and, crucially, containing the one lesson learned from 2018’s public scolding. The AI now introduces itself.
A feature in search of a user
So who is this actually for? The initial user base is tiny. It is a sliver of a fraction. The feature is launching as an early experiment, available only to a specific subset of Google’s customers in the United States. You need the new Pixel 11 phone to use it. That is the first hurdle. You must also pay for a Gemini subscription, a recurring fee for access to Google’s premium AI tools. This is not a generous gift to all Android users. It is a perk for a small, paying, and technologically committed group. The audience is minuscule.
The value proposition for that user seems clear enough. It is convenience. It is the outsourcing of a minor life annoyance. No one enjoys waiting on hold, listening to tinny music while trying to change a dinner reservation or check if a shop has a specific item in stock. This new system promises to take on that burden. The user types a command into their Pixel. The AI does the rest. It makes the call, it waits, it talks to a human, and it reports back, all while the user monitors the process through a live transcript. You save ten minutes. You avoid a frustrating phone tree. For some, particularly those with a strong aversion to phone calls, this will feel like a genuine upgrade. You pay your subscription. You get a robot butler to handle the tedious parts of your day.
But the real customer may not be the user at all. The real customer is Google. Calling this an ‘early experiment’ is the key. Experiments are designed to generate data, and every call initiated by a paying subscriber is a rich source of it. Every interaction with a baffled receptionist, every successful navigation of a company’s automated phone system, and every moment spent on hold is a lesson for the AI. This is not just a product. It is a vast, distributed training programme. The user pays a fee, and in return for some convenience they direct Google’s AI into a live training exercise with a real business.
Google gets paid twice. First, it gets the subscription money, a direct revenue stream that helps justify the immense cost of developing these complex AI models. Second, and far more importantly, it gets free, high quality, real world data to refine its core technology. The user is transformed from a simple consumer into an active, if unwitting, participant in the AI’s education. They choose the target. They define the task. The AI executes it and learns from the result, becoming more capable for the next user, and making the overall Gemini system a more valuable asset for Google. The user pays for the privilege of helping a trillion dollar company improve its products. The convenience is just the incentive.
The unpaid training partner
Now consider the other person. The one who picks up the phone. They are not a paying subscriber. They are not a willing participant. They are simply at work. Perhaps they are a junior employee at a local garage, trying to find a customer’s keys while the phone rings. They answer. A voice tells them it is a Google AI calling on behalf of a user. The employee has not opted into this. They cannot opt out. They are now, without consent, a training partner for one of the world’s largest corporations.
Their time is not free. Yet they are providing a free service to Google. The AI might be slow. It might misunderstand a question about whether a specific brand of tyre is in stock, forcing the employee to repeat themselves, to rephrase, to speak more simply for the benefit of the machine. All the while, a real human customer might be waiting on another line, or standing at the counter, their own patience wearing thin. This interaction is pure friction. It is a cost imposed by a distant technology company on a small local business. The user saves five minutes. The business loses five minutes. The transaction is not neutral.
This cost is not abstract. Every moment an employee spends patiently guiding an AI through a simple query is a moment they are not serving a present customer, stocking a shelf, or closing a sale. A tied up phone line can mean a missed booking from a real person who gets an engaged tone and decides to call a competitor instead. These small inefficiencies accumulate, representing a new, hidden operational tax levied by Silicon Valley. The small business owner is effectively subsidising Google’s research and development budget with their own payroll, providing the labour needed to iron out the kinks in a product they will never use and from which they gain nothing. The AI learns. The business pays.
The arrangement is profoundly unequal. A paying user in the United States directs their robot assistant to interact with a business that has no say in the matter. That business provides the real world, messy, human environment which Google needs to improve its AI, turning its employees into unpaid data labellers. The value flows in one direction only, from the local business straight to Google’s servers in Mountain View. Convenience for the phone shy user is bought at the direct expense of the person on the other end of the line. They are not a partner. They are a resource. They are the raw material.
An arms race for your local restaurant
The logical endpoint is an arms race. A quiet, digital one. It will be fought over the phone lines of local businesses. Business owners are not passive. They will not simply accept their staff becoming unpaid training assistants for a Silicon Valley giant. They will react. At first, they might just hang up. Soon, they will seek a technological defence, a system to filter out the robots, to guard the precious time of their human employees. This is how the escalation begins. The small business is forced to buy its own AI.
A new market will emerge. A market for counter AI. Systems will be sold to restaurants, hairdressers and mechanics, designed for one purpose. That purpose is to intercept Google’s Gemini. Imagine the scene. A customer’s Gemini, a product of vast server farms and years of research, calls a local pizzeria to make a booking. It is met not by a person, but by the pizzeria’s own cheap chatbot. The two machines then begin a slow, clumsy negotiation about a table for four at 7:30pm on a Friday. The human user, and the human pizza chef, are now mere spectators to this robotic back and forth. They watch a transcript. They hope for the best.
This creates a new digital divide. The bots may not understand each other. Google’s sophisticated AI, trained on the entire internet, may find itself unable to communicate with a simple system bought by a small business in Bolton. The booking fails. The tyre stock cannot be confirmed. The user receives a blunt ‘request failed’ message, their convenience suddenly obstructed by an invisible technological barrier. The user will blame the business. The business, now paying a monthly subscription for a defensive bot it never wanted, is caught in the middle. It is a new tax on existence. A toll for operating in a world shaped by Google’s choices.
Something vital is discarded in this process. There was always more to these brief calls than the exchange of information. A human conversation contains texture. It contains clues. You could gauge if a restaurant was busy and chaotic or quiet and intimate from the sounds in the background. You could build a tiny moment of rapport with the person who answered, a fleeting connection that might make you feel more welcome when you arrived. It was a small thing. A very small thing. But it was a human thing. By optimising this single interaction for the convenience of one person, we risk paving over these small, unscripted moments, replacing the messy and unpredictable reality of human contact with the sterile efficiency of two machines talking to each other in the dark. The convenience is for one. The cost is for all.
Sources. The Verge: Gemini can now call businesses for you so you don’t have to wait on hold. TechCrunch: Google tests letting Gemini call businesses for you.
Analysis. Drafted with AI assistance from the sources listed above and reviewed by an editor before publication. Jnews links to the organisations it writes about.
