Google testing 'Gemini Agent' upgrade that takes actions across apps, makes purchases autonomously
Google is testing a major upgrade to Gemini Agent, internally called "Remy," that can autonomously take actions on users' behalf including making purchases, sharing documents, and communicating with others. The experimental feature, available to Google AI Ultra subscribers, will monitor user preferences and handle complex tasks proactively across connected apps.
Google testing 'Gemini Agent' upgrade that takes actions across apps, makes purchases autonomously
Google employees are testing an upgraded version of Gemini Agent, internally named "Remy," that can autonomously perform actions including making purchases, according to internal documents reported by Business Insider and code found in Google app version 17.20.
What Gemini Agent does
The upgraded agent is described in internal documents as "a 24/7 personal agent for work, school, and daily life" that "can take actions on your behalf — not just answer questions or generate content." According to strings in the latest app code, it can:
- Communicate with others and share documents
- Make purchases on the user's behalf
- Access data from connected apps, personal context, chat history, uploaded files, and location
- Monitor for things that matter to users and handle complex tasks proactively
- Learn user preferences over time
The feature greets users with "What can I get done for you today?"
Current availability and warnings
Gemini Agent launched as an experimental feature with Gemini 3 in November 2025, available to Google AI Ultra subscribers. Google includes several warnings:
"It can make mistakes and expose data unintentionally. Be clear with your requests, and take care when asking it to do sensitive tasks. Supervise its tasks and actions in your dashboard. Gemini is not designed for tasks requiring professional knowledge, like for legal, medical, or financial actions."
The system saves browser and cookie data including login credentials and browsing history, though users can clear this data and manage settings.
Interface details
Users interact with the agent through conversations, with a dedicated section in the Gemini navigation drawer showing:
- Completed tasks
- In-progress tasks
- Tasks requiring user input
- Scheduled tasks
All tasks can be pinned, renamed, and continued later.
Privacy and control
The agent learns from user interactions, tasks, personal context, and uploaded files. Users can manage and delete activity through Gemini Apps Activity, turn off Personal Intelligence and Connected Apps, and manage personal context in Settings.
What this means
Google is positioning Gemini Agent as a direct competitor to autonomous AI assistants from Anthropic and others. The ability to make purchases and take actions across apps represents a significant escalation in AI agent capabilities, though Google's warnings about mistakes and data exposure suggest the technology remains experimental. The internal testing phase indicates this upgrade could ship to Google AI Ultra subscribers in coming months, though Google has not announced a public release timeline.
Related Articles
Google Gemini Live gains access to Memory and Connected Apps from past conversations
Google has updated Gemini Live to access past conversation history through Memory and Connected Apps. The feature, currently available in English in the US, allows the voice assistant to reference previous chats and information from YouTube, Workspace, Utilities, and image generation tools during conversations.
Mistral AI adds Deep Research agent, voice mode with Voxtral model to Le Chat
Mistral AI has released a major update to Le Chat, adding a Deep Research agent that generates structured research reports, a new voice input model called Voxtral, and Projects for organizing conversations. The update also includes multilingual reasoning powered by Mistral's Magistral model.
U.S. government orders Anthropic to halt exports of Mythos and Fable AI models, both now offline for one week
The White House ordered Anthropic to restrict exports of its Mythos and Fable AI models last Friday, citing national security concerns. Anthropic pulled both models offline within 90 minutes of the Commerce Department directive, marking the first major test of AI export controls.
GitHub details Qubot, internal Copilot-powered data analytics agent for plain language queries
GitHub has released technical details on Qubot, an internal analytics agent powered by GitHub Copilot that enables employees to query company data using natural language. The agent represents GitHub's implementation of AI-assisted data analysis for internal operations.
Comments
Loading...