AI Trend Research

← AI glossary

AI glossary · Advanced · Agents

A2A

A2A (Agent2Agent) is a common communication standard (protocol) that lets AI agents built by different companies or with different tools find each other, hand off tasks and exchange results.

About this term
Level
Advanced · Specialist terms that need the inner workings
Fields
Agents
Added
2026-10-08
In plain words It is similar to how people from different countries can send each other parcels if they all use the same international postal form. The similarity extends further: a postal form only defines how things are sent and received, and does not guarantee that the recipient will do the job well.

How it works

Each agent publishes a self-description called an "Agent Card." It lists what the agent can do, its endpoint address, how to authenticate and so on. The agent that wants to hand off work (the client agent) reads these cards, picks a suitable counterpart (the remote agent) and sends it a "Task." The remote agent reports its progress while processing the task and, when finished, returns outputs such as documents or data. Long-running tasks can continue with intermediate updates exchanged along the way.

How work is handed off with A2AUser request → Client agent; Agent Card → Client agent(Read and choose); Client agent → Remote agent(Send task); Remote agent → Client agent(Progress status); Remote agent → OutputsUser requestClient agentHands off the workAgent CardCapabilities andaddressRemote agentReceives the workOutputsDocuments and dataRead and chooseSend taskProgress status
How work is handed off with A2A The client agent reads Agent Cards, picks a counterpart and sends a task; the remote agent returns progress updates and outputs.

Why it matters

Companies often have a mix of agents built by different vendors, and if each one connects differently, they have to be wired together one by one. With a common standard like A2A, aligning once is enough for many agents to work together. This makes it easier to build multi-agent systems, in which several agents split up roles, in practice. Google first released it, and it is now run as an open project with participation from multiple companies.

Things to keep in mind

A2A is often mentioned alongside MCP. MCP is a standard for connecting agents to tools and data, while A2A connects agents to other agents, so the two are closer to complementary. The protocol only defines how agents talk; whether the other agent is trustworthy or its results are accurate must be checked separately. Handing work to an external agent can send information outside, so authentication and permission management also matter.

Example

For example, when a company's business-travel assistant agent receives a request from an employee, it can use A2A to send separate tasks to a flight-booking agent and a hotel-booking agent, then combine the two results into an itinerary. Even if the three agents are products of different companies, they can connect because they follow the same standard.

Related terms

AI agent Multi-agent MCP Tool use Agentic RAG