Google Gemini adds autonomous task management features

Photo: Arwahyn, AI-generated illustration
The Rundown
Google updated Gemini with a new Task mode to automate daily errands like placing phone calls, scheduling calendar dates, and paying utility bills.
The details
- Google rolled out the update for free across the US, Japan, Korea, Canada, Australia, and India using the Gemini 4 Hydrogen model.
- Users can trigger task mode through a toggle in the Gemini app, via RCS texting, or by forwarding emails to gemini@google.com.
- Tasks run inside isolated, sandboxed virtual machines managed by Google Cloud's Agent Compute Engine to prevent cross-account leaks.
- Gemini integrates with the Agent Payments Protocol (AP2) alongside Visa, Mastercard, and American Express to handle checkouts within user-defined budgets.
- During outbound calls to businesses, Gemini explicitly discloses upfront that it is an artificial intelligence.
The full story
Google has rolled out a major capability update to its Gemini application, transitioning the tool from a conversational assistant into an autonomous task agent. Powered by the Gemini 4 Hydrogen model, the feature is launching free of charge across several countries including the United States, India, Japan, Canada, South Korea, and Australia. Users can engage these capabilities by toggling from Chat to Task mode on their mobile devices, sending RCS messages, or forwarding relevant messages to a dedicated email address.
The updated assistant directly handles multi-step personal chores. It can locate nearby businesses on Google Maps, place simultaneous phone calls, wait through hold music, and book appointments directly onto a Google Calendar. The system also accesses past billing records via Google Password Manager integration to diagnose utility price spikes, compares products through Google Shopping Graph, and carries out checkouts using the Agent Payments Protocol (AP2) in partnership with major card networks.
To address user privacy and security risks, Google executes these background tasks in dedicated, sandboxed environments. Powered by Google Cloud's Agent Compute Engine, each assignment runs inside a hardware-isolated virtual machine tied strictly to the specific user account. The architecture includes prompt-injection defense layers and scanning through Google Safe Browsing. Crucially, Gemini pauses to request manual confirmation before executing critical actions like payment checkouts, while allowing users to set strict spending limits.
Google acknowledged that website navigation across the public internet remains an evolving challenge for personal assistants. Because many existing sites still lack open agent standards, Gemini occasionally completes intermediate steps and prompts users to click final confirmation buttons themselves. Additionally, the phone calling service faces regional constraints within certain US states. Google plans to expand web checkout reliability next by integrating the Universal Commerce Protocol into the assistant framework.
Why it matters
This development moves consumer artificial intelligence from text generation into real-world autonomous execution. By integrating directly with payment protocols and system-level applications like dialers and calendars, Google sets a major precedent for personal agent capabilities. It accelerates the transition toward standardized agentic commerce online.
What it means for you
While the direct phone calling and local checkout integrations are not fully available in Bangladesh, you can already test forwarding administrative emails to see how it organizes events. Freelancers managing foreign clients can observe these workflows to understand how personal agent infrastructure will alter routine operational tasks. Watch how AP2 and Google Workspace automations develop to prepare your own client-facing services.
Source: Google AI
Spot a mistake? Tell us on WhatsApp.