On September 24, 2026, Google announced "Call for Me," an experimental feature that lets its Gemini assistant place phone calls to local businesses on a user's behalf. Google bills Gemini Call for Me as an early preview: the AI dials from the user's own number, navigates automated phone menus, waits on hold, and talks through a task with whoever answers — while the user watches a live transcript and can take over at any point. It is the most visible consumer example yet of the industry's shift from AI that talks to AI that acts.
The launch arrives as agentic AI moves out of demos and into real transactions. Just this week, Amazon blocked Meta's Muse agent from its storefront even as Amazon opened its own seller tools to outside agents, and Mastercard rolled out Agent Pay credentials for AI shoppers. Phones that call businesses themselves fit the same pattern: agents are becoming intermediaries between people and the economy.
Duplex grew up: from scripted demo to general-purpose caller
This is not Google's first attempt at machine-placed phone calls. Back in 2018, the company demoed Duplex, a system that could call restaurants and hair salons to book reservations, complete with human-like pauses and filler words that briefly alarmed listeners who realized they had been talking to software. Duplex was narrow — it worked for a handful of scripted scenarios — and it quietly faded from headlines after the initial buzz.
Call for Me is built on a general-purpose language model instead of a scripted system. According to TechRepublic's reporting, the assistant dials directly from the user's cellular number, announces itself as an AI, and handles tasks like checking inventory, rescheduling a haircut, or booking patio seating at a restaurant. Gemini's speech capabilities have advanced considerably in recent months, which underpins the feature's ability to handle back-and-forth conversation rather than only scripted exchanges.
The wider range matters because the phone call is one of the last holdouts of analog friction. Booking engines absorbed the easy half of scheduling years ago, but plenty of businesses still run on "call us." Google's example flow is simple: tell Gemini, in the Gemini app, "Call the restaurant and book a table for two at 7 PM tomorrow." Afterward, the Pixel Phone app provides a summary, a full transcript, and a recording of the call.
Guardrails: disclosure, live transcripts, and hard no's
Google is keeping humans within reach. Users can watch the live transcript as the call unfolds and take over the conversation at any time, which changes the risk profile compared with fire-and-forget automation. According to a report from TPS, the feature grew out of a "Call for me" capability Google introduced last year — which had Gemini call local businesses to check product availability, but without the option for users to take over mid-call, a capability added in this iteration. When someone picks up, Gemini introduces itself as an AI calling on the user's behalf and discloses that the call is being recorded. It shares user-approved personal details — a name, a booking reference — only where the user granted permission.
The feature also has hard boundaries. Google has published no data on how often Gemini Call for Me completes calls without the user stepping in — a gap that matters for anyone evaluating delegated agency at scale. Gemini cannot complete financial transactions or share health information, and the user has to step in for those moments. Because the call is placed through the user's own number and cellular network, a business that blocks automated callers could also block the user's number — an interesting tension between spam prevention and delegated agency that phone carriers and regulators will eventually have to untangle.
Availability is deliberately narrow for now. Call for Me is in beta for users over 18 in the US, and it requires a paid Google AI subscription plus one of the newest Pixel 11 devices — Pixel 11, Pixel 11 Pro, Pixel 11 Pro XL, or Pixel 11 Pro Fold — enrolled in the Phone by Google beta. As Thought Catalog reports, Google says access will roll out gradually, so even eligible users may not get it immediately.
Why this launch matters for agents everywhere
Tying a headline AI capability to flagship hardware and a paid subscription is a classic Google move: it turns a risky, open-ended behavior into a measurable experiment. Voice calls are messy — accents, hold music, menu trees that hang up on silence — and Google wants real usage data before it commits to a full-scale launch. But the company has a long history of pushing Pixel-exclusive features to the broader Android ecosystem once they are proven out.
The broader significance is about where the agent economy is heading. This month alone has seen execution-verification tooling for agents (like Archipelo's Salmon runtime), one-time-use payment credentials for AI shoppers through Mastercard Agent Pay, and now AI that picks up the phone. Each piece closes a gap between what an agent can decide and what it can actually do.
The missing piece, as analysts keep noting, is accountability. A transcript exists, the AI identifies itself, the user can intervene — but nobody has published data yet on how often Gemini completes calls without the user stepping in. The verifiable-calling future depends on measurement, and right now the only numbers are Google's own.
Still, the direction is unmistakable. For years, the conversation was about generative AI — tools that answer questions and draft emails. The conversation now is about agentic systems that plan a multi-step task, choose the right tool, execute, and adjust when something goes wrong. One of those tools, apparently, is the telephone. Hold music finally has a worthy opponent.
Comments 0
No comments yet. Be the first to share your thoughts!
Leave a comment
Share your thoughts. Your email will not be published.