Google · 2026-05-19 · major
Google Ships Managed Agents in the Gemini API — One Call Spins Up an Ephemeral Linux Sandbox Where Antigravity Reasons, Calls Tools, and Runs Code
A new Gemini API endpoint provisions a sandboxed Antigravity agent on demand. Developers define behavior with AGENTS.md and SKILL.md markdown files, and sessions resume with preserved state across follow-up calls.
A single Gemini API call now provisions an ephemeral Linux sandbox with an Antigravity agent inside it.
What is it?
Managed Agents is a new capability inside the Gemini API that lets developers spin up production-grade AI agents without owning the sandbox infrastructure. Each call returns an agent — powered by the new Antigravity agent built on Gemini 3.5 Flash — running in an isolated, ephemeral Linux environment where it can reason, plan, call tools, execute code, and browse the web.
How does it work?
Agents are defined declaratively with two markdown files: AGENTS.md describes the agent's role and operating instructions, and SKILL.md files document individual skills the agent can call on. Each invocation provisions a fresh Linux sandbox with file management primitives. Sessions can be resumed across follow-up API calls with preserved state, and the harness is the same one Google ships in Antigravity 2.0, repackaged behind a single API call. The infrastructure builds on Google's Deep Research stack.
Why does it matter?
Building a production agent has meant running your own sandboxes or paying a third-party agent runtime. Google now offers that as a Gemini API primitive next to model inference, directly mirroring Anthropic's Claude Managed Agents and OpenAI's Codex platform. Defining behavior in markdown instead of orchestration code lowers the bar for teams that want a managed agent without bespoke infra.
Who is it for?
Gemini API developers, agent platform teams, enterprises comparing managed-agent offerings across labs
Try it
Preview rolling out today at ai.google.dev/gemini-api/docs/agents and via Google AI Studio.