Google I/O opens on May 19, and two weeks before the event some Google employees began testing an internal agent called Remy. Materials seen by 9to5Google describe it as a 24/7 digital partner, not a new model but a more active form of Gemini.
Remy is designed to bring verbs into Gemini: buying online with saved payment methods, sending files through Gmail, Drive and Docs, controlling a Mac, organizing files, operating software, monitoring topics that matter to the user and learning preferences over time.
The shift is from passive response to persistent work. Instead of answering one prompt and stopping, Remy would take a goal, plan steps, call tools, move across apps and ask the user only when a decision or uncertainty requires it.
The data access is unusually broad: chat history, connected apps, Personal Context, Personal Intelligence, uploaded files, location, plus browser credentials, cookies and browsing history. Chrome-level access is the line many agent platforms avoid.
Google can attempt this because it already controls much of the data and product surface. Claude lacks Gmail, Calendar and Drive; Google combines data, model and endpoint control.
Google's dogfooding warning is blunt: Remy can make mistakes and expose data unintentionally. It is not suitable for legal, medical or financial judgment, and autonomy, confirmations, permissions and auditing remain open questions.
The timing fits the wider market. Anthropic has pushed Claude Cowork, ServiceNow and Nvidia have Project Arc, and OpenAI has emphasized agent-driven systems. If Remy appears formally at I/O, it may matter more to users than another Gemini benchmark gain.
Sources: Google preps 'Gemini Agent' as your '24/7 digital partner' (9to5Google); CocoLoop, Google's Gemini May Be Getting a 24/7 AI Agent Called 'Remy' (eWeek); Google tests Remy AI agent for Gemini as focus turns to user control (AI News); Google develops Remy AI agent for Gemini (Let's Data Science)