Google is testing Remy, a new AI personal agent for Gemini that can take actions for users in work and daily tasks, according to AInews. The tool represents Google’s latest push to expand Gemini beyond simple chat-based responses into a more proactive assistant.
The testing is currently limited to a staff-only version of the Gemini app. An internal document describes Remy as a “24/7 personal agent” designed to turn Gemini into an assistant that can act on a user’s behalf. Two people familiar with the project confirmed that Google employees are testing Remy, though a Google spokesperson declined to comment.
This development matters because it signals Google’s serious intent to compete in the emerging AI agent market. While companies like OpenAI have gained attention with autonomous AI tools, Google has been more cautious about releasing agents that can act independently. Remy suggests the company is ready to take bigger risks in this space.
The report describes Remy as more advanced than Google’s existing agent features, including Agent Mode. It’s designed to integrate deeply with Google services, monitor user-relevant information, handle complex tasks, and learn user preferences over time. However, key technical details remain unclear, including whether Remy can act independently without user confirmation.
Google’s current Gemini ecosystem already connects with numerous services to complete user requests. The connected apps include:
- Google Workspace services (Gmail, Calendar, Docs, Drive, Keep, and Tasks)
- Third-party platforms like GitHub, Spotify, YouTube Music
- Communication tools including WhatsApp
- Smart home controls through Google Home
- Android device utilities
The focus on user control comes at a crucial time for AI agents. Google’s existing documentation covers various action types with different levels of user impact, from retrieving information to sending messages and controlling smart devices. The company has also established principles for AI agents, including well-defined human controllers, limited powers, observable actions, and planning capabilities.
Privacy and control features are central to Google’s approach. The Gemini Privacy Hub lets users review and delete activity, manage auto-delete settings, and control whether their data improves Google’s AI systems. Users can also manage app access and information they’ve asked Gemini to save.
Remy’s reported preference-learning function puts additional spotlight on memory controls. This capability would allow the agent to become more personalized over time, but it also raises questions about data handling and user privacy that Google will need to address before any public release.
The internal document describes Remy as a “dog-fooding” project, using the tech industry term for employees testing products before broader release. This approach mirrors how other major AI developments have rolled out, starting with internal testing before moving to limited public access.
The timing is significant as competition heats up in the AI agent space. The report compared Remy to OpenClaw, an AI agent that gained attention for autonomously replying to messages and conducting research. OpenAI CEO Sam Altman said in February that his company was hiring OpenClaw’s creator, showing how seriously major tech companies are taking autonomous AI development.
Google DeepMind CEO Demis Hassabis has previously discussed building a digital assistant, but the company hasn’t confirmed whether Remy will become a public Gemini feature. The lack of details about release timing or public availability suggests Google is still evaluating how to balance capability with safety and user control.
The broader implications extend beyond Google. As AI agents become more capable of taking autonomous actions, questions about oversight, transparency, and user control become more pressing. Google’s emphasis on user control and privacy features may set important precedents for how the industry approaches these powerful new tools.
