Google Unveils Gemini Spark AI Agent at I/O 2026

Google announced Gemini Spark, an always-on AI agent designed to handle routine tasks in the background while users focus on other work. The agent can write emails, track schedules, create updated study guides, and scan credit card statements for hidden fees. It runs continuously on virtual machines hosted in Google Cloud.

Spark connects to Workspace applications including Gmail, Docs, Sheets, and Slides. Google is also expanding support to third-party services through the Model Context Protocol, an open standard for linking AI models to external systems. Integrations planned for Canva, OpenTable, and Instacart will let the agent book reservations and manage design tasks.

Later this summer, users will gain the ability to interact with local files through the Gemini app on macOS. The system will request permission before carrying out high-stakes actions such as sending payments or emails. Gemini Spark begins limited testing this week and will reach beta availability for Google AI Ultra subscribers in the United States next week.

Google also updated its Antigravity coding tool with a new desktop application, command-line interface, and software development kit for building custom AI agents. Users will eventually be able to communicate with Spark through text and email, similar to existing agent platforms.

Google plans to add Chrome connectivity and live updates through a forthcoming interface called Android Halo.