Google has started rolling out Gemini Spark, a proactive AI agent that runs in the background and automates tasks across Google Workspace, third-party apps, and the open web. The agent is available in beta for Google AI Ultra subscribers in the US at $100 per month, according to 9to5Google.
Spark was previewed at Google I/O 2026 last week. It runs on Gemini 3.5 and uses Google’s Antigravity harness, a cloud-based execution layer that keeps the agent working even when the user closes their laptop or locks their phone.
How Spark Works
Spark operates through three components: Tasks (high-level goals like planning a trip), Schedules (time-based or event-based triggers), and Skills (reusable instruction sets that teach the agent how to use specific tools). Users can create custom skills using natural language, according to Thurrott.
Within Google Workspace, Spark can search email threads, draft replies, organize inboxes with labels, manage calendar events, create and edit Docs, build Sheets with formulas, and generate Slides presentations. Beyond Workspace, the agent has access to a remote browser for navigating websites, adding items to shopping carts, and interacting with web pages, according to 9to5Google.
Google stated the agent “keeps working in the background even when you close your laptop or lock your phone,” positioning Spark as a persistent, cloud-native execution layer rather than a session-bound chatbot interaction.
Limits and Guardrails
Users can run up to 15 tasks simultaneously. Scheduled tasks will not trigger if the task limit is reached, per Times of India. Spark always asks for user confirmation before performing sensitive actions, including sending emails or completing transactions.
Google said it is working on additional integrations with Canva, OpenTable, and Instacart. Custom sub-agents and the ability to text or email Spark directly are planned for future updates, according to Thurrott. A macOS desktop app is expected this summer.
The Agent Platform Race Tightens
Spark’s launch comes days before Microsoft Build 2026 (June 2-3), where Microsoft is expected to announce its own Windows-native agent capabilities alongside Nvidia-powered agent PCs at Computex. OpenAI’s Operator already handles autonomous web-based tasks. Google is betting that deep Workspace integration and persistent cloud execution give Spark an advantage over competitors that require the user’s device to stay active. For teams already running Google Workspace, Spark’s native access to Gmail, Calendar, Drive, and Docs eliminates the API plumbing that third-party agent frameworks require.