Google Enhances Gemini API Managed Agents with New Features

Google has announced significant enhancements to its Managed Agents within the Gemini API. These new features are designed to help developers build more reliable and production-ready AI agents, addressing common needs for enterprise applications Source.

Managed Agents in the Gemini API enable businesses to leverage AI for tasks like reasoning, code execution, and data management within a secure cloud environment. The latest updates expand these capabilities, making it easier to integrate AI into existing business processes.

Key New Capabilities Arrive for Gemini Agents

The updates focus on four critical areas, improving how AI agents manage tasks, interact with external systems, and handle security credentials.

Asynchronous Background Execution

Previously, long-running AI tasks could be challenging due to the need for a continuous connection. Google is introducing background: true for interactions, allowing tasks to run asynchronously on the server. This means an API call immediately returns an ID, which client applications can use to monitor progress or reconnect later. This capability is ideal for processes that don't require an immediate response but need to run reliably over extended periods Source.

For example, an agent could be tasked with analyzing large datasets or generating detailed reports in the background without tying up user-facing applications.

Remote MCP Server Integration

Businesses often have private databases and internal APIs. The new remote Model Context Protocol (MCP) server integration allows managed agents to connect directly to these internal systems without needing custom proxy middleware. This streamlines the process of giving AI agents access to proprietary data and services in a secure manner.

Agents can now combine built-in sandbox features, like Google Search or code execution, with custom tools that connect to an organization's MCP servers. This expanded connectivity enhances the agent's ability to perform complex, domain-specific tasks while adhering to security best practices Source.

Custom Function Calling

Developers can now add custom functions alongside the built-in sandbox tools. This feature enables step-matching, where built-in tools execute automatically on the server, while custom functions prompt the client application to execute local business logic. This allows for a hybrid approach, where AI agents can leverage both Google's powerful tools and an organization's specialized local functions.

This is useful for scenarios where an AI agent needs to perform an action using a specialized internal system that isn't publicly accessible through the API directly. The agent can 'call' a custom function, and the client application handles the execution of that specific business logic.

Network Credential Refresh

Managing access tokens and API keys, which often have short lifespans, is critical for secure operations. The Gemini API now supports refreshing credentials or rotating keys by passing an existing environment_id with a new network configuration. This means that updated security tokens can be applied without disrupting the agent's ongoing work, preserving its filesystem state and installed packages Source.

This ensures continuous, secure operation of AI agents, reducing downtime and administrative overhead associated with credential management.

Impact for Businesses

These updates transform managed agents into more versatile and robust workers. They can operate independently on long-running tasks, seamlessly integrate with internal business infrastructure, and maintain secure connections. For business leaders, this means AI can be deployed more effectively for automation, data analysis, and complex operational support, paving the way for more sophisticated AI-driven solutions.