Vertex AI’s New Upgrade: The Best Way to Build and Manage Multi-Agent Systems

In the near future, almost every enterprise will rely on “Multi-Agent Systems” — multiple AI agents collaborating across platforms to complete tasks. These agents can act on your behalf, possessing reasoning, planning, and memory capabilities, and can even think several steps ahead, collaborating across systems to handle complex tasks.

To build multi-agent systems, the key lies in three pillars: powerful models (such as Gemini 2.5), deep integration with enterprise workflows and data, and a robust operating platform. This is precisely the area where Google Cloud’s Vertex AI is focused.

Google Cloud today officially launched several new Vertex AI features to help you build AI agents in a more open way and deploy them with enterprise-grade control.

Agent Development Kit (ADK) Debuts

This is Google’s open-source development framework, which uses the same architecture as our internal Agentspace and Customer Engagement Suite. ADK allows you to create an AI agent with less than 100 lines of code. Through examples and expandable templates, you can quickly get started in Agent Garden.

You can:

• Precisely define the agent’s reasoning, collaboration, and decision-making processes.
• Interact naturally with agents using two-way audio/video streaming (beyond pure text input).
• Support multiple models, including Gemini and over 200 models in Model Garden (such as Claude, Meta, Mistral AI, etc.).
• Deploy on Cloud Run, Kubernetes, Vertex AI, or local debugging environments.

Agent Engine: Helping You Deploy Stably to Enterprise Environments

Agent Engine is a fully managed execution platform on Vertex AI, allowing you to go from development to deployment seamlessly, eliminating tedious infrastructure setup. Here are four supported items:

• Multi-framework (ADK, LangGraph, Crew.ai, etc.) and multi-model support.
• Short-term and long-term memory functions, allowing agents to remember context.
• Evaluation tools and Example Store to help you continuously optimize agent performance.
• Integration with Google Agentspace to extend AI throughout the enterprise.

Seamless Communication: Agent2Agent Protocol Officially Launched

One of the biggest challenges for enterprises is the difficulty of collaboration between agents from different vendors and architectures. To address this, Google Cloud has teamed up with over 50 industry partners to jointly launch the open Agent2Agent protocol. This protocol can:

• Allow agents from different platforms (regardless of framework or model) to communicate using a common language.
• Declare capabilities and coordinate interaction methods (text, forms, voice/video, etc.).
• Maintain high security and consistency.

Currently, companies including Box, Deloitte, Salesforce, UiPath, and ServiceNow have joined this initiative.

Connecting Agents to Your Enterprise Data and the Real World

Using ADK and the MCP (Model Context Protocol) standard, you can integrate agents with internal APIs, databases, and workflows without rebuilding or duplicating data. Supported resources include:

• Internal data sources like AlloyDB, BigQuery, NetApp.
• Over 100 pre-built connectors.
• Google Maps (updated with over 100 million map information daily) providing geographical context.

These integrations ensure that AI responses are based on your most familiar and accurate data, providing results that are close to actual needs.

Comprehensive Security Protection, Enterprise-Grade Trust Guarantee

For enterprise-scale AI adoption, security is a non-negotiable requirement. Using Gemini and Vertex AI, you can:

• Set content safety filters and system instructions to ensure AI responses comply with brand and regulations.
• Control agent access permissions (personal accounts or service accounts).
• Utilize VPC Service Controls to isolate and protect sensitive data.
• Set interaction protection (such as input validation, tool parameter restrictions, etc.).
• Use end-to-end tracing tools to monitor every step, including reasoning and tool execution.

This article references the Google Cloud Blog official website. Whether you want to learn about the latest cloud knowledge, or stay updated on the latest events and industry applications, please feel free to contact Microfusion Technology. We will bring you more AI cloud insights, so please pay close attention to our event information, and we look forward to seeing you at our events!