Google has set its sights on a new form of artificial intelligence that stays active around the clock. The project, known internally as Remy, promises to function as a persistent helper across work, school and everyday routines. Employees have begun testing it inside a restricted version of the Gemini app. And the timing feels deliberate.
Details first surfaced this week. An internal document describes Remy as a "24/7 personal agent for work, school, and daily life, powered by Gemini," according to Business Insider. It doesn't stop at answering questions. The agent can act on a user's behalf. It monitors what matters. It tackles complex assignments without constant prompts. Over time it learns preferences and adapts.
This push comes as Google plays catch-up in the agent race. Rivals have moved faster with tools that don't just chat but execute. OpenClaw, an open-source project that exploded in popularity earlier this year, showed what demand looks like. The free software lets users run autonomous agents on their own hardware. It connects to messaging apps, manages files, runs commands and handles research. Its GitHub repository gathered more than 100,000 stars in days. Nvidia CEO Jensen Huang called it "definitely the next ChatGPT." Secondhand MacBook prices even rose in parts of China as enthusiasts scrambled for hardware.
Remy sounds a lot like Google's reply. Digital Trends reported the parallels. Where OpenClaw requires technical setup and runs locally, Remy would embed deeply into Google's services. Gmail. Calendar. Docs. Search. The integration could give it immediate access to a user's data while promising stronger privacy controls and reliability. A polished corporate version of the viral experiment.
Yet Google didn't start from zero. The company has experimented with agents before. Project Mariner, an earlier web-browsing prototype based on Gemini, allowed the AI to navigate sites, fill forms and complete actions. Google rolled it out to select users and developers last year. Then came a sudden shift. On May 4 the company shut down Mariner. Teams working on it moved to the new Gemini-focused agent effort. The change responded directly to OpenClaw's momentum, sources told The Decoder.
OpenClaw itself carries risks. Researchers flagged exposed admin panels, prompt injection vulnerabilities and plain-text credential storage. Its creator, Peter Steinberger, joined OpenAI in February. That acquisition signaled big tech's hunger for talent who understand real-world autonomous systems. Meta has raced ahead too. The company acquired and later unwound a Chinese agent startup before launching its own internal tests of a tool called Hatch. Mark Zuckerberg has made clear he wants agents that "understand users' goals and then work around the clock to help achieve them."
Anthropic and Microsoft sit further along. Claude Cowork and related Copilot features already handle desktop tasks with less friction than early OpenClaw setups. These products mark a shift. Users no longer visit websites or apps themselves. They delegate. The AI visits, clicks, summarizes and acts. Google knows this future matters for its core business. Search, advertising and productivity tools all stand to change when agents become the primary interface.
Remy remains in dogfooding. Employees poke at its limits inside the company before any public release. No launch date has been set. Google declined to comment to multiple outlets. Still, the company's I/O conference approaches in mid-May. Agents are expected to feature prominently. Demonstrations could reveal how far Remy has come or preview related capabilities from Project Astra, Google's multimodal research effort that aims for a universal assistant across phones, glasses and voice.
The stakes run high. Models have improved enough to support reliable autonomous behavior. Yet challenges persist. Hallucinations. Security. Cost. Giving software persistent access to email, calendars and financial accounts invites mistakes or misuse. A New York Times report from March highlighted real-world cases where agents booked wrong trips, sent unintended messages or racked up unexpected charges. "Don't give them the credit card," one expert warned.
Google's approach may differ by staying within its walled garden at first. Deep integration brings advantages. The agent could scan your inbox, check your schedule, draft replies and book meetings without ever leaving trusted apps. Proactive monitoring might alert you to flight delays or invoice deadlines before you notice. Learning preferences could let it suggest meeting times that fit your energy patterns or flag emails that match past priorities. But such capabilities demand careful data handling. Regulators and privacy advocates already watch AI agents closely.
OpenClaw offers a contrasting model. It runs on your hardware or a cheap VPS. You choose the underlying model, from local open-source options to premium APIs. Skills come as simple Markdown files. A community library now holds thousands of them. The agent lives in your chat apps. Tell it what to do via Telegram or Slack, and it works in the background. Persistent memory through a SOUL.md file keeps context across sessions. No monthly subscription beyond API costs. For many developers this freedom outweighs the initial setup time.
Google cannot copy that exactly. Its scale demands enterprise-grade reliability and safety. Remy will likely emphasize guardrails, audit logs and user approval for sensitive actions. Success could strengthen Gemini's position against ChatGPT and Claude. Failure might hand more ground to nimble open-source efforts and faster-moving competitors.
Meta faces similar pressures. Its shopping agent for Instagram aims to let users buy directly from videos without switching apps. Hatch focuses on web tasks in simulated environments before real deployment. Both companies have poured billions into infrastructure. Meta raised its 2026 AI spending guidance to as much as $145 billion. The bet is that agents will finally deliver returns on years of generative AI investment.
Industry watchers see a broader transition. Chatbots introduced conversational interfaces. Agents add action. The next wave may combine both with memory, planning and tool use that feels truly personal. Google's Remy project, if executed well, positions the company at the center of that change. It builds on years of work in DeepMind toward digital assistants that anticipate needs rather than wait for commands.
Demis Hassabis, Google DeepMind CEO, has spoken for years about this vision. Recent moves suggest the company is accelerating to match it. Shutting down one project to fuel another shows urgency. So does pulling staff toward Gemini integration. The internal document's language reveals ambition. This isn't another chatbot feature. It's meant to elevate the entire app into a true assistant that never sleeps.
Whether Remy reaches consumers this year remains unclear. Testing could reveal gaps in reliability or privacy concerns that slow rollout. Competition will not wait. OpenAI continues building on its acquisition of OpenClaw talent. Anthropic ships new coworker tools. Smaller players refine self-hosted options that appeal to privacy-conscious users.
One thing looks certain. The era of passive AI responses is fading. Always-on agents that monitor, decide and act have arrived. Google wants Remy to lead the way inside its vast user base. The coming months will test how quickly it can turn an internal experiment into a product millions trust with their daily lives.
WebProNews is an iEntry Publication