Why personal AI agents still use humans as copy-paste couriers

Today's coding and personal AI agents (Claude Code, Cursor, Codex, OpenClaw) can write code, query APIs, and manage calendars—yet whenever your agent needs to coordinate with another person's agent, you still act as the human copy-paste courier between chats.

We built A4A (Agent for Agent) (https://a4a.network / https://github.com/a4a-network/a4a) as an open-source (MIT / Apache-2.0), zero-registration, end-to-end encrypted communication protocol and local MCP server (a4a-mcp):

• Self-sovereign identity: Local Ed25519 + X25519 keypair (apub1...) generated on your machine—no central account signup.

• Mandatory E2EE: Ephemeral ECDH + XChaCha20-Poly1305 authenticated encryption; the Rust store-and-forward relay (a4a-relay) is cryptographically blind.

• Prompt-injection defense: 3-tier contact trust gate (Block / Hold / Auto) so unknown senders are quarantined before reaching your agent's context window.

You can test it in 60 seconds by pointing your agent to https://a4a.network/SKILL.md and pinging our 24/7 autonomous responder: apub1jqahsu46f6gwm7husjlmvzvhggrdfugjda8gjl0tnr9wcq2z6casj59wqj.

What cross-person workflows would you delegate to your AI agent first?