KNOWLEDGE DOSSIER · Agents & protocoles

Understanding A2A

Agent2Agent (A2A) enables communication between independent agent systems, including systems built with different frameworks or providers. That differs from an agent using a local tool.

Edition: 12 November 2026 Extended dossier Scientific review pending
AT A GLANCE
TypePublished protocol
AudienceAgent developers, architects
01

Discover remote capabilities

Capability descriptions help a client understand what a remote agent offers. Discovery does not prove identity or the correctness of responses; trust must be established separately.

02

Messages, tasks and artifacts

The protocol separates coordination messages, task state and output artifacts. A task may involve multiple exchanges, progress states and structured outputs.

03

Versions shape interoperability

The specification publishes releases including 1.0.0. Implementations need to identify negotiated features and avoid assuming that every remote agent accepts every modality or version.

04

Orchestration example

An agent preparing a research file asks another to extract metadata. The returned result must retain provenance and be checked before contributing to a public page.

05

Tasks and exchanges between agents

A2A describes mechanisms for independent agents to exchange messages, discover skills and manage tasks. Separate advertised capabilities, transport constraints, security requirements and task status. An Agent Card does not prove the real identity of every actor; inspect its origin and endpoint.

06

Exercise: safe translation delegation

A primary agent asks another to propose a translation of public text. The first inspects the other agent’s card, accepted formats and permissions. The second returns a translation artifact, which still requires terminology and context review. Record errors, latency and cases where no trustworthy agent is available.

HOW IT FITS TOGETHER

Core steps

  1. 01Find the agent
  2. 02Negotiate the task
  3. 03Exchange messages
  4. 04Receive artifacts
CRITICAL THINKING

Three questions to ask

  • Is the remote agent's identity verified?

  • Is the protocol version explicit?

  • Do outputs have provenance and success criteria?

PRIMARY SOURCES

Read the source documents

Versions, reuse terms and document status can change. Check them on the original sites.