Google has announced Managed Agents support for the Gemini API, allowing developers to run autonomous AI agents inside secure cloud-hosted Linux sandboxes with a single API call. The feature was introduced through Google’s official developer blog during Google I/O 2026.  

According to Google, developers can now run the Antigravity agent directly through the Gemini API or build custom agents using their own instructions, tools, skills, and datasets. The system supports versionable configuration files like AGENTS.md and SKILL.md for defining agent behavior and capabilities.  

Google says Managed Agents can autonomously:

• Execute code

• Manage files

• Browse the web

• Run workflows

• Use tools inside cloud sandboxes

The company describes the feature as a major step toward production-grade autonomous AI systems.  

The launch is closely tied to Google’s broader “Agentic Gemini Era” strategy announced at I/O 2026, where Gemini 3.5 Flash, Gemini Spark, and Antigravity 2.0 are being positioned as foundations for long-running AI agents and workflow automation systems.  

Google says developers can start building managed agents immediately through the Gemini API and Google AI Studio.

• Official Announcement

• Gemini API Agents Documentation