Google employees test 'Remy' 24/7 personal agent; KPMG and others begin enterprise pilots

Google’s latest AI story is really two stories at once. One is consumer-ish — employees are testing an internal Gemini agent called Remy that aims to act on your behalf across work, school, and daily life. The other is enterprise — KPMG, UC Riverside, and Google Cloud itself are showing what agents look like when they touch real company systems instead of just answering prompts. The important change is not “AI got smarter.” It’s that AI is getting connected, permissioned, and governed. ### What is Remy, exactly? Remy looks like Google’s attempt to move Gemini from assistant to operator. Internal descriptions say it lives inside a staff-only Gemini app and works as a “24/7 personal agent” that can integrate with other Google services and take actions for users, not just generate text. That makes it sound less like a better chatbot and more like Google’s answer to the new wave of agent products that promise to handle messages, research, scheduling, and follow-through. ### Why is that a bigger deal than another chatbot? Because the hard part was never getting a model to sound useful. The hard part was giving it enough access to actually finish a task without turning enterprise security into a dumpster fire. Once an agent can see calendars, documents, internal systems, and workflow tools, it stops being a demo and starts becoming software that can change how work gets done — access, orchestration, and security controls. ### What are companies actually piloting? KPMG is one of the clearest examples. In late April, it expanded its Google Cloud alliance around Gemini Enterprise to build agents for complex work in regulated industries. One finance-focused agent, built with Google Cloud and integrated into Workday, is meant to help with month-end close tasks, anomaly review, and financial analysis. That is a very different target from a generic office copilot — it is narrow, process-heavy, and tied to a system of record. ### Why does UC Riverside matter here? Because it shows the same pattern outside corporate consulting. UC Riverside soft-launched a campus AI platform called The Grove on April 29. It gives early users Gemini chat, NotebookLM, and agents that can draw on university systems while staying inside existing campus account permissions. Basically, the university did not launch a magical all-knowing bot. It launched a closed, campus-connected assistant that inherits the access rules people already have. ### Why do permissions matter so much? Permissions are the whole game. If an agent can see everything, nobody serious will trust it. If an agent can only see what the user could already access, the security model becomes much easier to reason about. That also helps with audit trails, approvals, and compliance — especially in finance, healthcare, education, and other regulated settings where “the AI did it” is no longer the side issue to the main event as agents spread. ### So are broad personal agents the future? Maybe for consumers, yes. But inside organizations, the winning pattern looks narrower. Not one giant front-desk super-agent that touches everything, but lots of smaller workflow-native agents — one for close, one for campus knowledge, one for support triage, one for document prep. That is less flashy, but it is easier. ### What’s the real takeaway? The agent era is starting to look less like a sci-fi concierge and more like a permissions-aware coworker. Remy is the eye-catching part. But the more durable story is the boring one — agents are becoming useful when they inherit access rules, stay inside governed systems, and do one real job well.