What is Agent-to-Agent (A2A) Protocol?

The A2A Protocol establishes a standard for how autonomous AI agents communicate with each other across organizational boundaries. At its core, every participating business publishes an agent.json file at the /.well-known/ path of their domain. This manifest contains a structured description of the agent's capabilities (what it can do), its supported interaction patterns (synchronous request-response, asynchronous task delegation, or streaming), and its authentication requirements. When Agent A needs to fulfill a complex task, say, sourcing industrial components, it crawls potential suppliers' agent.json manifests, evaluates their capabilities against requirements, and initiates a multi-turn negotiation. The protocol supports three interaction modes: immediate (single request-response), deferred (task submission with polling), and streaming (real-time data feeds). Crucially, A2A implements a "Task" abstraction with states (submitted, working, input-required, completed, failed) that allows agents to manage long-running business processes. The verification layer uses ECDSA cryptographic signatures to ensure that agents can prove their identity and authority without exposing credentials. This is what enables zero-trust B2B transactions between organizations that have never interacted before.