[Say Goodbye to Disconnections and Sandbox Resets: Gemini API Managed Agents Introduce Background Asynchronous Execution and Direct Remote MCP Connections] According to monitoring by Beating, Google has announced several foundational infrastructure updates for Gemini API Managed Agents to enhance the stability of agents running in the cloud. Managed Agents are agents that operate within Google's isolated sandbox environment. Previously, developers faced three major challenges when deploying production-grade applications: long-running tasks were prone to HTTP timeout disconnections; accessing private data required setting up custom proxy servers; and updating authorization keys would reset the sandbox, causing the loss of installed dependencies and cloned git repositories. Four key adjustments directly address these challenges: 1. **Asynchronous Background Execution**: By passing `background: true`, tasks can now run asynchronously in the cloud. The API immediately returns a task ID, allowing the client to avoid maintaining a persistent connection and instead poll for status or reconnect to retrieve progress at any time. 2. **Direct Remote MCP Service Connection**: Agents can now securely communicate directly with remote Model Context Protocol (MCP) servers, enabling access to private databases or internal interfaces without the need for intermediary proxies. 3. **Hybrid Function Calls**: Sandbox tools run on the server side, while custom local functions will have their status marked as `requires_action`, delegating specific business logic to the client for local execution. 4. **Lossless Refresh of Network Credentials**: When developers rotate short-lived API keys or credentials, they can use `environment_id` to directly overwrite configurations while preserving the cloud sandbox's file system, cloned repositories, and installed dependencies. Managed Agents are now available in the API's free tier, and developers can directly invoke them using the `@google/genai` development toolkit. [Original Link]