Google’s Gemini Spark wants to run your digital life 24 hours a day: the new AI agent reads your emails, drafts your replies, and never needs you to keep your laptop open

Google unveiled Gemini Spark at its I/O developer conference on May 19, 2026 — a cloud-based AI agent that runs around the clock on dedicated virtual machines, with or without the user’s devices switched on. Built on Gemini base models and an agentic harness from Google Antigravity, Spark is designed to monitor inboxes, draft emails, manage documents, and handle multi-step tasks across the Google Workspace suite.

What Makes Spark Different from a Standard AI Assistant

Most AI assistants wait for a prompt. Spark was built to operate persistently on cloud infrastructure, executing tasks whether the user’s laptop is open or not.

Alphabet CEO Sundar Pichai described the product at I/O as an agent that acts “on your behalf and under your direction,” running on dedicated virtual machines at Google Cloud without relying on local hardware.

Reading, Drafting, and Managing the Inbox

Spark’s primary capability at launch is Gmail integration. Users can instruct it to monitor messages and prepare draft replies using information from Gmail threads, Google Docs, Sheets, and Slides.

Josh Woodward, VP of the Gemini App and AI Studio, demonstrated how Spark can write a status-update email by pulling relevant facts from documents already stored in the user’s account.

Sub-Agents, MCP Connections, and Third-Party Apps

When given a complex task, Spark can spawn sub-agents to handle different components in parallel. Google has not yet released detailed documentation on how permissions and approval controls will operate within that framework.

Beyond Google Workspace, Spark will connect via MCP to third-party providers including Adobe, Asana, Box, Canva, Dropbox, HubSpot, Spotify, and Wix, with more integrations planned.

How Users Interact with It

Users communicate with Spark by emailing a dedicated Gmail address, or by including it in text chains and email threads. Chrome integration lets the agent browse and interact with the web independently.

Google is also building Android Halo, a mobile interface for monitoring Spark’s live task progress. The company has said Android Halo will arrive later in 2026.

Pricing and Availability

Spark launched in limited beta for trusted testers at Google I/O, with a broader rollout to AI Ultra subscribers the following week. Access requires an AI Ultra subscription, which Google has restructured into two tiers: $100 and $200 per month.

That pricing excludes subscribers on Google’s lower-cost AI Pro plan. Initial availability is limited to the United States.