Google dropped a big update for its Managed Agents in the Gemini API, focusing on making them more production-ready. The biggest change is background execution—instead of keeping an HTTP connection open for long tasks (which is fragile), you can now set background: true and get back an ID to poll or stream progress later. That’s huge for building reliable apps.
They’ve also added direct support for remote MCP (Model Context Protocol) servers, so you can connect agents to private databases or internal APIs without writing custom middleware. Plus, you can mix custom function calling with built-in sandbox tools for more flexible workflows.
Another neat feature: network credential refresh. If your access tokens expire, just pass a new config with the same environment ID, and the sandbox keeps its files, packages, and repos intact.
Why it matters: This turns managed agents from simple demos into real asynchronous workers that can handle complex, long-running tasks without blocking your app—a critical step for deploying AI agents in production environments.
