Last updated:
A networked topology in which AI agents discover, communicate with, and delegate work to one another through a shared infrastructure layer.
An agent mesh applies the principles of service mesh architecture (popularized by tools like Istio for microservices) to agentic AI systems. Rather than agents calling each other point-to-point, the mesh provides a managed communication layer handling agent registration, discovery, authentication, message routing, and telemetry. From a governance perspective, the mesh becomes the natural enforcement boundary: action policies, audit trails, and human-in-the-loop controls apply at the mesh level rather than being re-implemented inside each agent. The terms agent mesh and agent fabric are often used loosely; mesh tends to emphasize the inter-agent communication topology, fabric the underlying runtime substrate.
Real world example:
A consumer goods company runs separate agents for procurement, supplier risk, and contract review. Rather than wiring direct integrations between them, the agents communicate through an agent mesh that authenticates each request, logs which agent invoked which, and lets the governance team throttle or halt agent-to-agent traffic when a policy is violated.
