Google unveiled Gemini Spark at its I/O developer conference, positioning the 24/7 personal AI agent as an evolution from standard chatbot to active task executor.

The agent runs on Gemini 3.5 and integrates with Google Workspace apps including Gmail, Docs, and Slides. Users can train Spark to handle recurring workflows like extracting critical deadlines from email or summarizing lengthy message threads.

Spark supports complex multi-step automation. For example, it can analyze meeting notes across chats and emails, generate polished reports in Google Docs, and draft accompanying emails for distribution.

The service launches to testers this week before reaching Google AI Ultra beta users in the US next week. Google prices its AI Ultra plan at $250 monthly.

Third-party integrations expand capabilities

Beyond Workspace, Spark connects to Canva, OpenTable, and Instacart at launch. Google plans additional partner integrations in coming weeks.

The company emphasizes user control through opt-in activation and granular app permissions. Spark requests approval before "high-stakes actions like spending money or sending emails," according to Google.

Upcoming features include text and email sending capabilities plus browser automation. Google plans to bring Spark to its desktop Gemini app this summer, enabling file access and computer-based task execution.

The launch positions Google against agentic AI competitors like Anthropic's Claude and emerging workflow automation platforms. Google's integration advantage lies in its Workspace ecosystem, where many enterprises already manage daily operations.

Spark represents Google's broader push into autonomous AI agents that move beyond conversational interfaces toward actionable business automation.