Interagent: A Coordination Protocol for Multi-Agent LLM Systems

[YOUR VOICE] The Claim

Every multi-agent framework on the market automates the coordinator role. Interagent makes the opposite bet: the human operator is the coordination mechanism, and the protocol exists to make that role sustainable at scale.

The Mechanism

The Interagent Protocol governs coordination across heterogeneous AI agents — Claude Code, Gemini CLI, Codex, Cowork — running in separate runtimes with no shared memory or state. The protocol provides:

1. Entity profiles — each agent has a declared identity, capabilities, and authority boundaries
2. Structured memos — communication between agents follows a format that can be audited and replayed
3. Operator routing — the human decides what information goes where, not an orchestration engine
4. State documents — each project maintains current-state, roadmap, and decisions docs that serve as ground truth

MISSING — Protocol specification details, message formats, routing rules

MISSING — The ia CLI tool interface and commands

The Evidence

MISSING — Operational metrics: 11 entities, 44 projects, coordination overhead measurements

MISSING — Failure mode catalog (feeds into the Failure Mode Watchlist)

[YOUR VOICE] Implications

MISSING — Why human-in-the-loop coordination is an architectural choice, not a limitation.

Open Questions

• At what agent count does the operator become the bottleneck?
• Can the protocol support partial automation (operator delegates routine routing)?
• How does the protocol evolve as agent capabilities improve?

Reference Documents