Executive Summary
Google has introduced Managed Agents in the Gemini API, a new feature that allows developers to run and customize powerful AI agents in a secure, Google-hosted cloud sandbox with a single API call. This service, powered by a new base agent named "Antigravity," abstracts away the complexity of infrastructure and sandbox management. Developers can now focus on defining agent behavior and skills using simple markdown files, significantly simplifying the process of building and deploying production-grade agents that can reason, use tools, and execute code.
Key Takeaways
* Managed Environment: Provides an isolated, ephemeral Linux sandbox where agents can execute code, manage files, and browse the web.
* "Antigravity" Base Agent: The experience is powered by a new agent called "Antigravity," which is built on Gemini 3.5 Flash.
* Declarative Customization: Developers can create custom agents by extending the base agent with their own instructions and skills defined in simple markdown files (`AGENTS.md`, `SKILL.md`).
* Stateful Sessions: The environment is persistent across API calls within a session, allowing an agent to resume tasks with all files and state intact.
* Availability: The feature is rolling out in preview in the Gemini API and is accessible via the Interactions API and Google AI Studio. A private preview is also available for enterprises.
Strategic Importance
This launch lowers the barrier to entry for building and deploying sophisticated AI agents, positioning Google's Gemini platform as a more accessible and developer-friendly choice for creating complex, tool-using AI applications.