Multi-Agent Systems: When One Agent Isn’t Enough

Supervisor–worker, peer handoff, and debate — when a second agent genuinely earns its tokens, what the multi-agent tax actually costs, and the failure modes (deadlock, divergent state, telephone-game briefs) that only exist once there is more than one context.

Content current as of 2026-09.

Lessons

1. The shapes: supervisor–worker, peer handoff, debate
2. When a second agent genuinely helps
3. The multi-agent tax, itemized
4. Failure modes only multi-agent systems have
5. The default: one agent, until the context objects