Announcing the Agent2Agent Protocol (A2A)
Source: Google Developers Blog Authors: Rao Surapaneni, Miku Jha, Michael Vakoc, Todd Segal Date: 2025-04-09
Core idea
A2A proposes an open interoperability protocol for agents from different vendors or frameworks, centered on capability discovery via Agent Cards, task-oriented communication, artifact exchange, and long-running task status updates.
Key claims
- Agent collaboration needs an explicit peer protocol, not only tool APIs or shared prompt conventions.
- Capability discovery, task lifecycle tracking, artifact passing, and message parts are distinct protocol concerns.
- Multi-agent ecosystems become more useful when cross-vendor delegation is standardized.
Harness takeaway
For a multiplayer harness network, A2A is a plausible peer-delegation layer: agents should be discoverable, describable, and able to exchange task objects and artifacts without pretending they all run inside one harness.