The AI Agent framework is taking shape! Google open-sources A2A, will "MCP + A2A" become the future standard?

Last night, Google open-sourced the first standard agent interaction protocol at the Google Cloud Next 25 conference — the Agent2Agent Protocol (A2A for short).

A2A will fundamentally break down system silos and qualitatively change the capabilities, cross-platform functionality, and execution efficiency of agents, supporting mainstream enterprise application platforms such as Atlassian, Box, Cohere, Intuit, Langchain, MongoDB, PayPal, Salesforce, SAP, ServiceNow, UKG, and Workday.

In simple terms, this A2A interaction protocol is reminiscent of Google's initiative years ago with over 80 companies to develop the Android system, as more than 50 well-known companies have already joined in the first batch. As more companies join, it will greatly enhance the commercial value of A2A and promote the rapid development of the entire agent ecosystem.

At this conference, agents became the focus. In addition to A2A, Google also followed OpenAI's lead by open-sourcing the Agent Development Kit (ADK), the internal testing tool Agent Engine, and a new Agent marketplace. Below, the "AIGC Open Community" will interpret the key points of A2A for everyone.

What is A2A

A2A is an open protocol that provides a standard interaction method for agents, enabling them to collaborate with each other, regardless of the underlying framework or vendor.

For example, a large e-commerce company uses multiple enterprise platforms and services. Atlassian is used for team project management, Box for file storage and sharing, Salesforce for customer relationship management, and Workday for human resource management.

Previously, agents on these platforms could not communicate freely. Now, with the A2A protocol, these enterprise platforms can securely and freely automate interaction data.

A2A Case Demonstration

When designing the protocol with partners, Google followed five key principles. First, A2A focuses on enabling agents to collaborate in their natural, unstructured modes, even if they do not share memory, tools, and context. Google is enabling true multi-agent scenarios rather than restricting agents to being a single tool.

Second, the protocol is built on existing, popular standards, including HTTP, Server-Sent Events (SSE), JSON-RPC, etc., which means it is easier to integrate with the existing IT stack that enterprises already use daily.

For example, an e-commerce company routinely uses the HTTP protocol to handle web data transmission and utilizes JSON-RPC to pass data instructions between the front and back ends. After introducing the A2A protocol, the company's order management system can interface with the A2A protocol via HTTP to quickly obtain logistics data updates provided by relevant intelligent agents There is no need to go through the trouble of rebuilding complex data transmission channels; it can easily integrate into the existing IT architecture, allowing various systems to work together more smoothly.

Third, A2A is designed to support enterprise-level authentication and authorization, and at launch, it has parity with OpenAPI's authentication scheme. This is quite user-friendly and does not exclude OpenAI; using the A2A protocol allows for quick authentication, secure data retrieval, and ensures the security and compliance of data transmission, preventing the risk of data leakage.

Fourth, Google designed A2A to be flexible, capable of supporting a range of scenarios from quick tasks to in-depth research that may take hours or even days (when human involvement is required). Throughout the process, A2A can provide users with real-time feedback, notifications, and status updates.

For example, in a research institution, researchers use Agents under the A2A protocol for new drug development-related studies. Simple tasks, such as quickly retrieving existing drug molecular structure information from a database, can be completed and fed back to researchers within seconds. However, for complex tasks, like simulating the reactions of new drug molecules in the human body, it may take several days.

During this time, the A2A protocol continuously pushes simulation progress to researchers, such as how many steps have been completed and current issues encountered, allowing researchers to stay informed, as if they have an assistant reporting on work progress at all times.

Fifth, the world of Agents is not limited to text; therefore, A2A supports various modalities, including audio, images, and video streams.

How A2A Works

The working principle of A2A is achieved by facilitating communication between the client Agent and the remote Agent. The client Agent is responsible for formulating and conveying tasks, while the remote Agent takes action based on these tasks to provide the correct information or perform the corresponding operations. In this process, the A2A protocol has several key capabilities.

First, Agents can promote their capabilities through "Agent Cards." These "Agent Cards" exist in JSON format, allowing the client Agent to identify which remote Agent is best suited to perform a specific task.

Once the appropriate remote Agent is identified, the client Agent can communicate with it using the A2A protocol to assign the task.

Then, task management is an important aspect of the A2A protocol. Communication between the client and remote Agents revolves around completing tasks. The protocol defines a "task" object, which has its own lifecycle.

For some simple tasks, they may be completed immediately; while for some complex, long-term tasks, Agents can communicate with each other to maintain synchronization on the task completion status. When the task is completed, its output is referred to as a "artifact." In addition, A2A also supports collaboration between Agents. Agents can send messages to each other, which can include contextual information, replies, artifacts, or user instructions. This way, Agents can work better together to complete complex tasks.

Finally, A2A also has the capability for user experience negotiation. Each message contains "parts," which are complete content fragments, such as generated images.

Each part has a specified content type, allowing the client and remote Agents to negotiate the correct format needed, and clearly include negotiations for user interface capabilities, such as iframes, videos, web forms, etc. This way, A2A can provide the best user experience based on user needs and device capabilities.

Which companies have joined A2A

What is most surprising is that A2A gained the favor and participation of a large number of well-known companies right after its release, including Accenture, Boston Consulting Group, Capgemini, Cognizant, Salesforce, Deloitte, Oracle, HCL Technologies, Infosys, KPMG, SAP, McKinsey, PwC, and more than 50 other companies that people often hear about.

The first batch of companies to join A2A

Several technology companies have also joined the A2A protocol. For example, Atlassian is known for its powerful team collaboration tools Jira and Confluence. By joining A2A, Atlassian can enable its tools to seamlessly collaborate with various Agents, further enhancing team productivity and collaboration capabilities.

Box is a company focused on enterprise-level cloud storage and content management services, providing solutions that help businesses securely store, share, and manage files. Through the A2A protocol, Box can integrate its services with Agents to achieve more efficient content management and automated workflows.

Intuit, a well-known financial software company, offers QuickBooks and TurboTax, which are widely used for financial management and tax processing. Through the A2A protocol, Intuit can enable its software to collaborate with Agents, achieving more automated financial processes and more efficient tax handling.

MongoDB is a company that provides high-performance, open-source NoSQL database solutions, with its databases widely used for data storage and management in modern applications. Through the A2A protocol, MongoDB can integrate its database services with intelligent Agents to achieve more efficient data management and automated data processing In fact, by now everyone should be clear that Google wants to unify the chaotic landscape of Agents and create a brand new execution and interaction standard, which is much more powerful than the recent MCP!

AIGC Open Community, original title: "AI Agent Major Changes! Google Open Sources A2A, Overnight Changing Agent Interaction"

Risk Warning and Disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial conditions, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article are suitable for their specific circumstances. Investing based on this is at your own risk