Google’s Agent2Agent interoperability protocol aims to standardize agentic communication

Interaction between AI agents slowly starting the stands for agents agents.

In the past few months, at least two agents have a reciprocal standard appeared: Anthropical Model Context Protocol (MCP) and AqNTCYFrom a guided collective Cisco. The importance of agents – especially in various frames and extensive language models (LLMS)Talking to each other and get a more full picture of the place where an enterprise has earned is vice as a new protocol.

Today, Google The agent opens a new research protocol named A2AGENT or A2A hopes that this will be standard within the industry.

Atlassian, box, coida, intuit, ligchain, mongodb, sap, sap, sap, servyenow, ukg and working days, agent2agent imperfection For agents and AI applications.

In an exclusive interview, Rao Surapani, Rao Surapani and Google Cloud’s Business Applied Platform and Director Venturebeat said that A2A facilitates agents with various qualifications and information nodes in the necessary context.

“Everyone has a certain specialization, because the information owner node or logic node, or the existing user database is directed to this particular position,” said Surapani. “You expect these frameworks to develop a very specialized carefully. If I am a customer and placing this many platforms and many frames, I do not want to make many chairs.”

Surapani is to build A2A for the fact that Google’s operation is “the ability to interact with an enterprise ready, reliable and reliable.

Building on existing standards

A2A facilitates the communication between Google, the customer agent and a remote agent. The customer agent develops the task from the end user and communicates and moves in a remote agent.

In a separate blog post, Google, A2A, said he depends on several major opportunities built on the protocol.

• Discovery: Agents can “advertise their capabilities” via the JSON format, so the customer agent can assign the best remote agent to complete a case.
• Task Management: Ensuring communication between agents is focused only to complete the inquiries and determines the life for the tasks.
• Collaboration: Send messages to messages, artefacts (performance performance) or instructions).
• User experience negotiations: Reading agents to identify content types and formats.

Surapani said that Google has prepared A2A as an open protocol, ie a larger open source community can contribute to the A2A project and offer code updates.

“We open it as an effort based on a society and the right source is open,” he said. “There is a management board around it, but we really want to be managed by the community.”

In the development of A2A, Google is aimed at working in cases where they do not share in natural, unstructured modals, memory, tools and context. The protocol is also based on existing standards such as HTTP and JSON, so it is easier to integrate with existing technological stacks and are valid as standard.

The rise of confusion protocols

Of course, the only interaction in the A2A market is not a protocol. Agntcy, Cisco, Langchain, Galileo, Llamaindex and a scare staff, aims to create a standard contact between agents. Langchain, a partner in agent2agent, has developed Agent Protocol. Microsoft has updated autogenous frames to help coat.

On the other hand, many companies, including Microsoft, have already hugged MCP. Google added support to MCP through the new agent development set. Surapane, A2A ensured that it will be parallel with MCP.

“We see MCP and A2A as complementary opportunities,” Surapani said. “The road we look at the agent2AGENT is in a higher layer of abstraction to ensure that applications and agents talk to each other. Therefore, think of MCP as a layered pile for tools and information.”

Surapani did not close the door on a possible cooperation with another consortia working on the agent’s interaction protocols. He said that A2A is always open to new members and the protocol will be a live code that is constantly updated based on community offers and needs.

“We will look at how to adapt all the protocols,” Surapani said. “There will always be a protocol with a good idea and we want to understand how all these good ideas will bring.”

There is confusion

Organizations and AI companies agree to the world’s desire Run in many EU models It means that it is more than a model that manages all, so agents will be built in different languages ​​and frames.

However, the full agent ecosystem requires agents with agents of other companies. But it’s easier than doing it. Industrial standards often take time to participate and make purchases from a large part of companies and demand.

A2A, MCP or Agntcy hopes to succeed in creating a standard way for all AI agents, or should be massive adoption and placement, no matter what frameworks are set.

Surapani admitted that there will be more than 50 partners working in A2A that adoption is still not in the fall.

“All of these protocols will develop, especially the growing way, and we will find new use and scenarios, as well as the growing way,” he said.