A2A (Agent-to-Agent)
A2A (agent-to-agent) refers to protocols and patterns for autonomous AI agents to communicate, delegate work, and coordinate with one another directly, rather than only interacting with a single human user or model.
Definition
A2A (agent-to-agent) refers to protocols and patterns for autonomous AI agents to communicate, delegate work, and coordinate with one another directly, rather than only interacting with a single human user or model.
Why agents talk to agents
Complex tasks increasingly involve multiple specialized agents — a planner that decomposes a goal, workers that execute sub-tasks, and reviewers that check results. A2A communication lets one agent hand work to another, share context, and combine outputs. As fleets grow, agent-to-agent interaction becomes the norm rather than the exception, which raises a governance question: when Agent A delegates to Agent B, whose permissions apply, and who is accountable for the resulting action?
Governing delegation
Multi-agent delegation can quietly amplify privilege if a low-trust agent can borrow a high-trust agent's reach. Cordum addresses this with delegation tokens that carry scoped, time-limited authority and support cascade revocation, plus dispatch-time re-verification so a revoked or expired delegation cannot still act. Every delegated action is evaluated against policy and recorded in the audit trail, so an A2A chain remains accountable end to end rather than becoming an anonymous relay of permissions.
Frequently asked questions
What is the governance risk in agent-to-agent systems?
Delegation can amplify privilege: a less-trusted agent may gain reach by having a more-trusted agent act on its behalf. Scoped, time-limited delegation tokens with revocation and per-action policy checks keep each step accountable.
How does A2A differ from MCP?
MCP standardizes how an agent calls tools and resources. A2A concerns how agents coordinate with each other. Both need governance — over tool calls in MCP's case, and over delegated authority in A2A's.
Related reading
Govern your AI agents with Cordum
Cordum is the agent control plane: policy-before-dispatch enforcement, human approvals, and a tamper-evident audit trail for autonomous AI agents.