At its annual I/O developer conference, Google introduced Gemini Spark, a new agentic personal assistant built on the company’s Gemini foundation models and an agentic framework called Google Antigravity. Alphabet CEO Sundar Pichai described Spark as the next step in digital assistants, capable of handling long‑horizon tasks with minimal user oversight while acting under the user’s direction.

Spark operates continuously on dedicated virtual machines hosted on Google Cloud, meaning users do not need to keep their devices powered on for the agent to stay active. It comes with out‑of‑the‑box integrations for Gmail, Google Docs, and the broader Google Workspace suite, allowing users to reach Spark simply by sending an email to a dedicated address. Through Chrome, the agent can browse the web, and on Android devices its progress can be monitored via the new Halo system.

Practical demonstrations showed Spark drafting status‑update emails by pulling relevant information from a user’s inbox, documents, spreadsheets, and presentations. Small businesses are already testing the tool to monitor customer inquiries and ensure no message goes unanswered. Like other agentic platforms, Spark can connect to additional services through the Model Context Protocol, with Google planning to expand those integrations in the coming months.

Currently in internal testing, Gemini Spark is slated for release to Google AI Ultra subscribers beginning next week. The launch signals Google’s push to embed proactive AI agents directly into everyday productivity tools, leveraging its deep access to users’ email and Workspace data to deliver a seamless, always‑on assistant experience.