Google Launches Managed Agents in Gemini API, Eliminating Infrastructure Complexity for Developers
Key Takeaways
- ▸Google opened up its managed agent infrastructure to developers, allowing custom agent creation beyond Deep Research
- ▸Agents run in isolated Linux sandboxes with persistent state, enabling complex reasoning, code execution, and web browsing
- ▸Developers define agents using markdown configuration files (AGENTS.md, SKILL.md) rather than complex orchestration code
Summary
Google has launched Managed Agents in the Gemini API, a new feature that abstracts away infrastructure complexity by enabling developers to build and deploy custom AI agents without managing isolated sandboxes or orchestration code. Powered by the Antigravity agent built on Gemini 3.5 Flash, managed agents can reason, call tools, execute code, browse the web, and manage files in ephemeral Linux environments provisioned with a single API call. Following Deep Research as its first managed agent launched in December, Google is now opening its agent harness to developers who can define custom agents using simple markdown files (AGENTS.md and SKILL.md).
Each agent interaction creates or receives an isolated environment where state and files persist across follow-up calls, allowing sessions to resume with full context intact. The feature rolls out today in preview via the Gemini API and Google AI Studio, with enterprise support available in private preview on the Gemini Enterprise Agent Platform. By removing the need to manage complex infrastructure scaffolding, Google is positioning itself to capture developers seeking production-grade agent solutions without significant operational overhead.
- Single API calls provision remote environments that support stateful multi-turn interactions
- Rolling out today in preview; enterprise support in private preview via Gemini Enterprise Agent Platform
Editorial Opinion
Google's decision to democratize agent infrastructure is a strategic move that could accelerate enterprise adoption of AI agents. By hiding infrastructure complexity behind a simple markdown-based abstraction, Google lowers barriers for developers who lack distributed systems expertise. The real value lies not just in the Antigravity agent itself, but in the foundation it provides for building a competitive ecosystem of custom agents—a direct response to OpenAI's push in agent development.
