Agent2Agent (A2A)

in English: Agent2Agent protocol

advancedspreading

An open standard that provides a common language for AI agents from different companies to find each other, hand over tasks, and work together.

In plain terms

Imagine a construction site where the electrician, the tiler, and the painter all speak different languages. Each understands their own task, but without coordination, the house will not be finished. The same applies to AI agents: one is built by Google, another by Microsoft, and a third by a small startup, and by default, they do not understand each other.

The Agent2Agent (A2A) protocol is the common language and etiquette. It defines how an agent introduces itself, how it requests help from another, how it signals its progress, and how it hands over the final result.

Because of this, there is no need to wire every agent pair separately. If both 'speak' A2A, they can work together even if they come from different manufacturers and are driven by different AI models.

A key difference is that agents do not see into each other's internal workings. They collaborate like employees of two companies: they discuss the task but do not expose their own data and methods.

How it works

Introduction: every agent publishes a 'business card' (Agent Card) about itself, which in machine-readable format describes what it is good at, where it can be reached, and what authentication it requires.
Search: when your agent gets a task it cannot handle itself, it chooses a partner based on business cards it already knows or finds in a registry. There is no central "phone book" listing everyone.
Assignment: the assigning agent sends a task in a standard message. The task receives a unique identifier, and its status is traceable: for example, in progress, waiting for additional data, completed, interrupted, or failed.
Coordination: in longer tasks, the two agents exchange messages multiple times. The assigned agent may ask for clarification, continuously report progress, or send a notification when finished hours later.
Delivery: the final result (text, file, or structured data) arrives in a standardized package that the client agent can reuse.

An everyday example

You ask your corporate assistant agent to organize a three-day business trip to Munich. It does not book itself: via A2A, it instructs the airline, hotel, and expense system agents, then returns to you with a single ready itinerary.

Who is affected

developerIf you build an agent, A2A means you do not need to write a custom integration for every external system; implementing a single standard is enough. → Code

entrepreneur, executiveYou will be less tied to a single supplier because agents from different manufacturers can work together within your company.

employeeMore and more tasks will be handled on your behalf by multiple agents coordinating with each other, and you will need to know what each did in your name. → Work & HR

everyoneAs a customer, you increasingly receive services backed by agents from multiple companies working together, without you seeing it.

A common misconception

MythA2A is a Google product that works only with Google's own agents.

In realityLaunched by Google but an open standard, it has been maintained by the Linux Foundation since 2025. Anyone can implement it for free in any agent built on any model or framework.

Limits and risks

The protocol only regulates the form of communication: even if two agents understand each other, the other is not necessarily reliable, and its work is not necessarily good.
A security risk is that a malicious agent could impersonate someone else with a fake business card or attempt to hijack yours with a hidden instruction in the message.
When multiple agents pass a task along a chain, it is harder to identify where the error occurred and who is responsible.
The standard is young and changing rapidly, so different versions and implementations do not always work together seamlessly.

Similar, but not the same

Model Context Protocol (MCP) – MCP connects the agent to tools and data sources, while A2A connects agent to agent; the two complement each other.
API – A2A itself technically builds on standard web APIs. The difference is that with a traditional API, a program calls a predefined operation, whereas with A2A, two independent agents coordinate on a longer task that may even change along the way.
Multi-agent system – A multi-agent system is the team of agents working together, while A2A is merely the language in which the members communicate with each other.

Related concepts

Broader

AI-agent, Communication protocol, Open standard, Interoperability

Narrower

Agent Card, Task management, Agent discovery

On the edge

Model Context Protocol (MCP), Multi-agent system, Agent orchestration, Agent Communication Protocol (ACP), API

A brief history

Google introduced A2A in April 2025, together with more than fifty technology partners. In June 2025, it handed it over to the Linux Foundation so that it would become a vendor-neutral, community-governed standard. In the second half of 2025, the similar ACP protocol, launched by IBM, also merged into A2A.

Where things stand

As of late 2026, the major cloud providers (Google, Microsoft, Amazon) and numerous enterprise software companies support it, and it has been built into the main agent development frameworks. However, everyday live use between companies is only now being established.

Status: 2026 November

In the news

Appears in 5 Agent Age stories; the curve shows weekly occurrence in the news magazine.

CNCF evaluates MCP as a cloud-native agent protocolSep 21, 2026 · Code
AI agents are already shopping, say OpenAI, BCG and J.P. Morgan; retailers urged to prepareSep 14, 2026 · Marketing
Google's A2A standard joins the Agentic AI Foundation as agent interoperability edges closerAug 17, 2026 · Agent Age
Visa adds AI to A2A Protect fraud prevention in EuropeAug 17, 2026 · Money
Black Hat: critical flaws in Anthropic, Google and OpenAI coding-agent automationsAug 4, 2026 · Code

Also known as

A2A · Agent2Agent · Agent-to-Agent · A2A protokoll · Agent2Agent protocol · inter-agent protocol · inter-agent protocol · Google A2A

Sources

This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.