Google Launches Gemini Spark: A 24/7 AI Agent That Wants to Make You Ditch OpenClaw

Google’s Gemini app has more than 900 million monthly users. Up until now, most of them used it as a very smart chatbox. That changes today.

At Google I/O 2026, the company announced Gemini Spark, a personal AI agent that doesn’t wait to be asked. It manages tasks across your apps around the clock, flags things that need your attention, and finishes jobs in the background while you’re busy doing literally anything else, including sleeping.

Spark runs on Gemini 3.5 and is built on Google’s Antigravity harness—the same agent infrastructure behind the company’s own internal tools. Unlike OpenClaw, Spark lives in the cloud, on dedicated Google virtual machines, so you don’t need your phone unlocked or your laptop open for it to keep going.

Out of the box, Spark can handle recurring tasks and do most of the everyday tasks you have handled to your Openclaw or Hermes agent.

It can automatically scan your monthly credit card statements for new or hidden subscriptions. It can monitor your kid’s school emails, pull out important deadlines, and send you and your partner a daily digest. It can pull raw notes from a meeting scattered across your Gmail and Docs, write them up into a clean document, and send the follow-up email to kick off the next phase of a project.

You can also teach it custom skills—essentially programming it with your preferences so it handles specific workflows the way you want them handled.

None of this was a complete surprise. Last week, developers digging through the Gemini Android app’s latest beta build found the “Gemini Spark” name—previously labeled internally as “Gemini Agent”—along with a welcome screen that described the same use cases Google announced today. Some internal references also called the project “Remy.”

The leaked code also included a disclaimer Google is keeping in the official release: Spark “may do things like share your info or make purchases without asking.” The company says it’s designed to request permission before high-stakes actions, but users are advised to supervise it.

What’s connected—and what’s coming

At launch, Spark works natively with Gmail, Google Docs, Slides, and the broader Google Workspace suite. Starting today, it also gains MCP connections to Canva, OpenTable, and Instacart—meaning it can start doing things like booking a restaurant or placing an order, not just drafting the message about it.

Decrypt wasn’t able to test the agent right now to see if it supports custom MCP configurations, because it’s still in limited access.

Over the summer, Google plans to add the ability to text and email Spark directly, create custom sub-agents, and let it operate your local browser. A macOS desktop version is also coming, which will extend its reach to local files.

The broader AI agent wave has been building for months—open-source tools exploded and self-improving agents started going viral on GitHub. Spark is Google’s answer to all of it: no terminal, no setup, no GitHub stars required. Just 900 million existing users, a deep integration with tools they already rely on, and an agent that gets to work the moment you turn it on.

Trusted testers get access this week. U.S. Google AI Ultra subscribers—whose plan Google is cutting from $250 to $200 a month—get beta access next week.