Google just revealed what it’s calling the most capable version of its Gemini AI yet. Gemini Spark, announced at the I/O 2026 conference on May 19, is a personal AI agent that manages email, calendar scheduling, and complex workflows around the clock, even when your devices are off.
What Gemini Spark actually does
Spark runs on the Gemini 3.5 model and operates inside what Google calls the Antigravity development environment. Rather than living on your phone or laptop, the agent runs on dedicated virtual machines within Google Cloud. That’s the key architectural difference here: it doesn’t need your device to be awake.
Users get a dedicated Gmail address for their Spark agent. You can email it tasks, and it executes them across Google’s product suite, including Gmail, Google Docs, and other Workspace applications. Progress tracking happens through something called Android Halo, a system that lets users monitor what their agent is doing in real time.
Gemini Spark is currently in testing. Google expects to make it available to Google AI Ultra subscribers starting next week.
