Google's Gemini Can Now Phone Businesses and Book Appointments on Your Behalf
The 'Tasks' feature, free in six countries, revives Duplex-style calling inside the Gemini app — while the company opens its Genie 4 world model to developers through the API.

Google has given its Gemini app the ability to place phone calls to businesses on a user's behalf — waiting on hold, booking the appointment and adding it to the calendar — as part of a free feature called Tasks rolling out in six countries, the company said this week.
Tasks, described in a Google Research post on November 20, positions Gemini as a "personal AI collaborator" that can execute multi-step real-world jobs. When a task requires calling a business — because no online booking is available — Gemini dials the number itself using technology descended from Duplex, the voice-calling system Google first demonstrated in 2018. The app discloses at the start of each call that the caller is an AI.
The feature is free across the United States, Japan, South Korea, Canada, Australia and India, though calling is excluded in five U.S. states whose consent laws Google has chosen not to test, according to an AI Tools Recap briefing. The assistant waits through hold queues, completes the booking and drops the result into Google Calendar.
Gemini dials the number itself, waits through hold queues, completes the booking and drops the result into Google Calendar.
The launch extends a week of agentic-AI escalation across the industry. Apple is hiring engineers for what a job listing describes as an AI agent that can code, browse the web and control apps, and Signal this week moved to block screenshots by default specifically to keep AI agents — including Gemini and Siri — from reading private messages, CNET reported.
Google paired the Tasks announcement with the developer release of Genie 4, its world model, through the Gemini API. Genie 4 generates continuous 720p interactive environments at 24 frames per second with synchronized native audio, starting from a text prompt or reference image and accepting live keyboard input and mid-session text prompts. Google has not published pricing — the number, one analyst noted, on which the model's economics will hinge.
Rivals are not standing still. Tencent published the weights of its Hy4 flash model on Hugging Face — roughly 224 billion parameters with about 12 billion active per token and a million-token context window — underscoring how quickly open-weight alternatives are closing on the features American labs monetize.


