Google has shipped four new capabilities for Managed Agents in the Gemini API, enabling developers to build AI systems that handle complex, long-running tasks without constant human oversight. The updates include background execution, remote Model Context Protocol (MCP) servers, custom function calling, and automatic credential refresh. All features are available immediately through the free tier of the Gemini API.

What Are Managed Agents and Why Do They Matter?

Managed Agents are AI systems within Google's Gemini platform designed to perform tasks autonomously. Think of them as digital assistants that can work in the background, make decisions, and interact with external tools and services. The new updates expand what these agents can accomplish, making them more practical for real-world applications where tasks take time or require connecting to multiple systems.

Google is making these tools available through its free tier, removing cost barriers for developers who want to experiment with agentic AI. Previously, developers faced limitations when building agents that needed to handle extended operations or authenticate with external services.

What Are the Four New Capabilities?

Google has introduced features designed to make Managed Agents more powerful and flexible for production use. Here is what developers can now do with the updated Gemini API:

• Background Execution: Agents can run long-running tasks asynchronously on the server by setting a "background: true" parameter, allowing them to work on complex problems over time without requiring instant completion.
• Remote MCP Servers: Developers can connect agents to remote Model Context Protocol servers, expanding the range of external systems and data sources that agents can access and interact with.
• Custom Function Calling: Agents can now execute custom functions tailored to specific use cases, giving developers more control over what actions their agents can perform.
• Credential Refresh: The system now handles automatic credential refresh, solving the technical challenge of authentication tokens expiring during long-running operations.

How to Build With Gemini Managed Agents

• Start with Free Tier Access: Developers can begin using Managed Agents immediately through the free tier of the Gemini API without requiring a paid subscription.
• Connect External Systems: Use remote MCP servers to link agents to databases, APIs, and other services that provide the data and functionality your agent needs.
• Design Long-Running Workflows: Leverage background execution to build agents that handle multi-step processes, such as data processing pipelines or complex research tasks that take hours or days.
• Implement Custom Logic: Write custom functions that define the specific actions your agent can take, ensuring it operates within your desired parameters and business rules.

What Else Is Google Shipping for Developers?

Beyond Managed Agents, Google has expanded its developer tools in other areas. The company recently added the ability to import projects directly from GitHub into Google AI Studio, allowing developers to bring existing code into the platform and continue iterating there. This feature makes it easier for teams already using GitHub to integrate Google's AI tools into their existing workflows.

Google AI Studio now includes a streamlined import process where developers can click a button to import a GitHub repository, and the platform automatically transforms it into a format compatible with its runtime environment. Developers can then deploy the project or continue building within AI Studio.

Where Do Gemini Models Stand in Performance Benchmarks?

Across various AI capability rankings, Gemini models continue to compete in different performance tiers. According to recent benchmark data, Gemini 3.1 Flash ranks among the fastest models for text-to-image generation, with Nano Banana 2 Lite, a lightweight variant built on Gemini 3.1 Flash, debuting at number five on the Artificial Analysis Text to Image Leaderboard. Meanwhile, Gemini 3.1 Flash TTS (text-to-speech) ranks in the top tier of speech synthesis models, though it was recently surpassed by SpeechifyAI's Simba 3.2 on the Artificial Analysis Speech Arena Leaderboard.

For coding tasks, Gemini Code Assist appears on lists of leading AI coding tools alongside competitors like Claude Code, GitHub Copilot, and Cursor. The diversity of Gemini model sizes, from Nano to Ultra variants, reflects Google's strategy of offering options for different computational budgets and latency requirements.

These updates position Gemini as a platform with multiple entry points for developers, from free tier experimentation with Managed Agents to specialized models optimized for specific tasks like image generation, speech synthesis, and code completion.