Google Announces Autonomous AI Agent Gemini Spark, Operates Even When Devices Are Off

Stunning Announcement at Google I/O 2026—The Autonomous AI Agent “Gemini Spark” Arrives

On May 20, 2026, at its developer conference “Google I/O 2026,” Google officially announced the personal AI agent “Gemini Spark.” This is not merely an evolution of a chatbot but a next-generation AI assistant that acts autonomously under user instructions to handle complex digital workflows. Its most significant feature is its ability to run 24/7 in the cloud, even when a user’s smartphone or PC is turned off. The concept of an AI proactively executing tasks while the user is asleep opens up new possibilities for boosting personal productivity.

Core Technology of Gemini Spark—Gemini 3.5 and the Antigravity Framework

Gemini Spark is built upon Google’s latest AI model, “Gemini 3.5,” and a dedicated agent development framework called “Antigravity.” Antigravity is described as the foundational technology enabling AI agents to autonomously judge and execute tasks by linking multiple tools and services. This allows Gemini Spark to “take actual actions based on instructions,” moving beyond the “answering questions” limitation of traditional AI assistants. Specifically, it operates in a dedicated virtual machine environment on Google Cloud, allowing tasks to continue independently of the user’s device state.

Background Task Execution Even When Devices Are Off

A groundbreaking aspect of Gemini Spark is that tasks continue running in the background even if a user closes their laptop lid or locks their smartphone screen. It supports both Android and iOS, with all task statuses synchronized in real-time across devices. For example, the AI can consistently handle multi-step processes, from checking business trip schedules to creating spreadsheets for expense reports. Users can later check progress via a dedicated dashboard and trace how the AI acted.

Deep Integration with Google Workspace—Automating Complex Workflows

Gemini Spark seamlessly integrates with Google Workspace tools like Gmail, Google Docs, Google Sheets, and Google Drive. Impressive examples were demonstrated at the announcement. For instance, if an event RSVP email arrives in Gmail, Gemini Spark automatically analyzes its content and updates an attendance spreadsheet. Furthermore, it sends reminder emails to those awaiting a reply and notifies the user as the response deadline approaches. The AI completes such cross-application tasks as a single, unified workflow.

Additionally, a feature called “Ghostwriter” learns from the user’s past writing to assist in document creation, matching their tone and style. For example, when automatically drafting a weekly report, it can generate natural text reflecting the user’s personality rather than just filling a template. Moreover, the “Planner” feature allows users to batch-assign multiple tasks, and the AI will formulate and execute an optimal plan in sequence.

Expansion with External Services—Building an Ecosystem via MCP Protocol

Beyond Google Workspace, Gemini Spark also connects with external third-party tools. The key to this is a standard protocol called “MCP (Model Context Protocol).” This enables integration with services like Canva, OpenTable, and Instacart, allowing automation of life-related tasks such as design creation, restaurant reservations, and grocery ordering. MCP functions as a common interface for agents to safely call external APIs, with the number of supported services expected to grow.

In the future, an “Agent Tab Browser” feature is planned to run Gemini Spark directly within the Chrome browser. This would allow the agent to handle browser-based tasks like navigating websites to gather information or filling out forms.

Safety and User Control—Confirmation Required Before Key Actions

Because the autonomous agent acts proactively, safety and transparency are emphasized. Gemini Spark is designed to always seek user confirmation before executing high-impact actions, such as adding or canceling reservations or transactions involving money. Furthermore, tasks running in the background can be monitored in real-time via a dedicated application dashboard, and paused or canceled as needed. The ability for users to oversee and control the AI’s actions is a critical point for ensuring trust.

Release Schedule and Future Outlook

Gemini Spark will first be made available to testers. For the general public, it is planned to be rolled out gradually to subscribers of the “Google AI Ultra” subscription service. Google positions Gemini Spark not just as a tool but as a “personal agent” supporting users’ digital lives, with future integration into voice interfaces and augmented reality (AR) environments also in view.

This announcement represents a symbolic case of the paradigm shift of AI from “dialogue” to “action.” While it promises to dramatically enhance personal productivity, the balance between AI autonomy and human control is likely to become a major topic of discussion. Google aims to realize “cooperative AI” where agents function as an extension of humans, and Gemini Spark is positioned as the first step toward that goal.