Google launches Gemini Spark, a 24/7 autonomous AI agent that works even when your devices are offline

Google unveiled Gemini Spark at its I/O developer conference on May 19, 2026 — a persistent, cloud-based AI agent designed to handle multi-step tasks across Gmail, Google Workspace, and third-party apps around the clock, without requiring users to prompt each step. The product has been rolling out in beta to Google AI Ultra subscribers in the United States since late May.

What's new

Gemini Spark represents a shift in how Gemini works: rather than answering questions on demand, it operates as an autonomous background agent that acts on behalf of users and reports back. Google describes it as transforming Gemini "from an assistant that can answer your questions into an active partner that does real work on your behalf."

Key capabilities at beta launch:

• Persistent cloud operation: Spark runs on Gemini 3.5 on Google's servers and "keeps working in the background even when you close your laptop or lock your phone"
• Google Workspace integration: Native connections to Gmail, Google Calendar, Drive, Docs, Sheets, and Slides; Spark monitors email threads, compiles meeting notes, and manages calendar events autonomously
• MCP ecosystem: Third-party tool integrations via the Model Context Protocol, including enterprise tools such as Microsoft SharePoint, OneDrive, and ServiceNow, as well as Adobe, Dropbox, and Uber
• User-directed automation: Users define Tasks, Skills, and Schedules to automate recurring workflows without writing code; the agent learns preferences over time
• Email interface: Users can text or email Spark directly as a task-delegation channel
• Human-in-the-loop safety: The agent "is designed to ask you first before performing high-stakes actions like spending money or sending emails"

Gemini Spark is available in two contexts: a consumer version for Google AI Ultra subscribers ($100/month and $200/month tiers launched at I/O 2026), and a business version through the Gemini Enterprise app rolling out to Workspace enterprise customers. Chrome and macOS support are planned for summer 2026. International availability has not been announced.

Context

Gemini Spark arrives as autonomous agent products have become a distinct competitive battleground. Anthropic launched Claude Managed Agents in public beta in April 2026, offering server-side agent execution with secure sandboxing via API. OpenAI has been expanding operator-configurable ChatGPT deployments. Microsoft has embedded autonomous Copilot agents into Microsoft 365 at the enterprise tier.

Spark runs on Google Antigravity's agentic harness — Google's internal framework for long-horizon agent tasks — paired with Gemini 3.5. The enterprise version operates in what Google describes as "a fully managed, secure runtime on Google Cloud, meaning you get enterprise-grade security without ever having to manage the underlying infrastructure."

The MCP protocol support is notable: it positions Spark as an orchestrator of the broader software stack rather than a Google-only assistant. By supporting existing enterprise connectors (SharePoint, ServiceNow), Google is targeting deployments where the data is not in Workspace.

Why it matters

Gemini Spark marks a concrete step toward AI agents taking real-world, persistent action inside users' software stacks rather than producing outputs that humans then act on. Granting an autonomous agent access to Gmail, Calendar, and third-party systems — with authority to draft and send communications, modify schedules, and execute multi-step workflows — represents a qualitative expansion of AI autonomy in everyday work.

Google's home-field advantage is significant: Workspace is already the daily productivity layer for hundreds of millions of users. Spark does not require users to adopt a new interface. The key product question the beta will answer is whether users will delegate meaningful inbox and calendar access to an autonomous agent — and whether the trust that requires can be built.