Agent2Agent — A2A
An open protocol standardizing communication and coordination between AI agents. Proposed by Google in 2025; migrated under the Linux Foundation as an independent project in 2026.
Definition
A2A has three primitives. (1) Agent Card: an agent declares its capabilities in JSON. (2) Task: the work unit agents exchange, with a lifecycle. (3) Transport: JSON-RPC 2.0 over HTTPS plus Server-Sent Events.
As of 2026, 150+ organizations have endorsed it (Microsoft, AWS, Salesforce, SAP, ServiceNow, Workday, IBM, and others). The v1.0 stable release adds multi-protocol support, enterprise multi-tenancy, and modernized security flows.
MCP connects agents to tools; A2A connects agents to each other. They are complementary and almost always deployed together.
Lemma Oracle implementation
Lemma layers verification onto A2A-resident agents through three hooks: (1) pin each agent's identity and capability declaration as a commitment; (2) record Task execution history as an audit trail; (3) verify inter-agent authority delegation in ZK.
Financial, public-sector, and regulated workloads need to be able to answer, after the fact, "which agent did what, with which authority." Binding the verification layer directly onto A2A's Agent Card and Task structures is how that becomes routine.
When A2A exchanges generate payments, the combination of x402 + Trust402 closes the loop into a single verifiable chain across coordination, payment, and audit.