Google developing ‘Remy’ personal Gemini agent to act across Gmail and apps

Google looks like it’s building the next version of the AI assistant — not a smarter chatbot, but a software operator. The reported project is called Remy, and the big idea is simple: Gemini should stop waiting for prompts and start doing things across the apps people already live in. That means email, calendars, files, design tools, chat, and code repos. If this ships anywhere close to the internal pitch, the product category changes a bit. ### What is Remy? Remy is the reported codename for a personal Gemini agent Google employees are testing internally. The pitch is that it works like a “24/7 personal agent” for work, study, and everyday tasks, inside Google’s employee Gemini app. The important shift is the verb — act, not just answer. It’s meant to operate across Gmail, Calendar, Drive, Slack, GitHub, and Figma, which turns Gemini from a text box into something closer to an executive assistant with API access. ### Why is that different from normal Gemini? Regular assistants mostly wait. You ask for a summary, a draft, a list, a rewrite. An agent is supposed to chain those steps together and then touch the tools where the work actually happens. So instead of “write me a follow-up email,” the system can read the thread, check the calendar, pull the file, draft the note, and possibly and error handling suddenly matter a lot more. ### Why those apps? Because they map to actual knowledge work. Gmail and Calendar cover communication and scheduling. Drive holds the documents. Slack holds the side-channel decisions. GitHub holds code and issues. Figma holds product and design context. Once one agent can move across all of that, it starts to look less like a feature and more like a work layer sitting above the software stack. That’s why this is strategically bigger than “Gemini gets another integration.” ### Is Google alone here? Not even close. OpenAI just introduced workspace agents in ChatGPT that run in the cloud, can be shared across teams, and are built for long-running workflows across company tools. So the race is no longer just whose model writes better prose. It’s whose assistant can safely connect to the most useful systems, remember the right context, and keep working when the user closes the tab. ### What changed this week? Google also rolled out persistent custom instructions for Gemini in Google Docs on May 4. That sounds small, but it’s actually the same design philosophy in miniature. Instead of re-explaining your tone, format, and preferences every time, you set rules once and Gemini keeps them in mind. That’s exactly the kind of memory-and-rules scaffolding a broader agent needs. ### Why are people watching May 12 and I/O? Because Google has already scheduled an Android-focused event for May 12, one week before I/O 2026, and both are expected to center heavily on Gemini. Even if Remy itself doesn’t get named onstage, these events are the obvious place for Google to show more agentic workflows, deeper Android hooks, and the breadcrumbs. ### What’s the hard part? Reliability. A chatbot that hallucinates is annoying. An agent that books the wrong meeting, edits the wrong file, or messages the wrong person is a much bigger problem. The winning products here won’t just be the most capable. They’ll be the ones that make delegation feel safe — with clear permissions, reversible actions, and enough memory to be useful without becoming creepy. ### Bottom line Remy matters less as a rumor than as a signal. Google seems to be pushing Gemini toward the same destination as everyone else — an assistant that lives across your tools, remembers your preferences, and does multi-step work on your behalf. If that becomes real, “using AI” stops meaning chatting with a model and starts meaning managing a digital coworker.