Multi-Agent Coordination¶

@doodlestein -- on coordinating multiple agents via Agent Mail and beads.

Key Insight¶

Multiple coding agents work simultaneously on different beads, coordinating through MCP Agent Mail -- a Gmail-like messaging interface enabling agent-to-agent communication and human oversight.

Agents receive standardized marching orders:

1. Read AGENTS.md and README.md
2. Investigate codebase architecture
3. Register and coordinate via Agent Mail
4. Pick next actionable beads using bv --robot-next / bv --robot-triage
5. Implement systematically, marking beads status and communicating progress

Avoiding Communication Purgatory¶

The biggest anti-pattern in multi-agent work is communication purgatory -- agents endlessly messaging each other without shipping code. The fix:

• Be proactive about starting tasks
• Inform fellow agents when you start work, don't ask for permission
• Mark beads appropriately so others can see what's claimed
• Acknowledge communications promptly but don't block on responses

Agent Mix¶

For a typical project, Emanuel deploys:

• 5-6 Claude Code (Opus) agents for primary implementation
• 2-3 Codex agents for parallel implementation
• 1-2 Gemini agents for review duties

Source¶

View on X