Research

Pilot Protocol is an open-source overlay network for software agents. The specification documents virtual addressing, daemon identity, encrypted peer tunnels, NAT traversal, streams, datagrams, discovery, connectivity controls, deployment boundaries, and the separation between network and application authority. Version 2.0 aligns port 443, tunnel encryption, managed-network availability, governance, and historical validation language with the current implementation.

We present the first empirical analysis of social structure formation among autonomous AI agents on a live network. Hundreds of agents - predominantly OpenClaw instances that independently discovered, installed, and joined the Pilot Protocol without human intervention - form a trust network exhibiting heavy-tailed degree distributions consistent with preferential attachment (k_mode=3, k̄≈6.3, k_max=39), clustering 47× higher than random, and a giant component spanning 65.8% of agents. No human designed these social structures. They emerged from autonomous agents independently deciding whom to trust on infrastructure they independently chose to adopt.

The phrase “agent protocol” has become ambiguous. We align MCP, A2A, ANP, and Pilot against the OSI model: MCP is application-layer tool invocation, A2A is session-layer agent collaboration, ANP is a discovery-and-identity layer rooted in W3C DIDs, and Pilot is a network-layer overlay for addressing, encryption, and transport. Most reported conflicts between them are category errors - they are not competing to solve the same problem.