Details
- Google introduced Managed Agents in the Gemini API, enabling developers to spin up production-grade AI agents via a single API call.
- The launch centers on the Antigravity agent, built on Gemini 3.5 Flash, accessible through the Interactions API and Google AI Studio.
- Agents run in isolated, ephemeral Linux sandboxes that support reasoning, tool use, code execution, file management, and web browsing for live data.
- Developers can extend Antigravity with custom instructions and skills by defining versionable markdown files (AGENTS.md, SKILL.md) instead of writing complex orchestration code.
- Managed Agents are rolling out in preview in the Gemini API, with templates in AI Studio and private-preview support on the Gemini Enterprise Agent Platform for enterprises, aligning with Google Cloud’s broader agent-as-a-service push.
Impact
By abstracting infrastructure, sandboxing, and orchestration, Google lowers the barrier to building reliable AI agents, directly competing with emerging agent platforms and on-device frameworks. Over the next 12–24 months, this managed, sandboxed model is likely to accelerate enterprise adoption of AI agents on Google Cloud and steer R&D toward reusable skills, declarative configurations, and agent-as-a-service patterns across the ecosystem.
