9.7.1 Multi-Agent Roadmap: Roles, Messages, Owner

Multi-Agent is a division-of-labor mechanism, not several chatbots talking. Use it only when role separation, parallel work, cross-checking, or specialist collaboration is worth the coordination cost.

See the Collaboration Cost First

The key question is: does the benefit of splitting work exceed the cost of messages, repeated context, conflicts, and final merging?

Run a Role Boundary Check

Every role needs one responsibility and one output. Keep one owner for the final decision.

agents = {
    "researcher": "collect evidence",
    "editor": "rewrite content",
    "reviewer": "check beginner clarity",
}

final_owner = "reviewer"

print("agent_count:", len(agents))
for name, job in agents.items():
    print(f"{name}: {job}")
print("final_owner:", final_owner)

Expected output:

agent_count: 3
researcher: collect evidence
editor: rewrite content
reviewer: check beginner clarity
final_owner: reviewer

If two roles produce the same output, merge them. If nobody owns the final decision, the system will drift.

Learn in This Order

Step	Read	Practice Output
1	When to use Multi-Agent	Write when a single Agent is better
2	Common patterns	Compare supervisor-executor, pipeline, debate, expert committee
3	Communication	Define message format, shared state, and handoff rule
4	Coordination	Track owner, queue, conflict rule, and aggregation
5	Practice and risks	Measure cost, loops, duplicated work, and role overreach

Evidence to Keep

Keep this page’s proof of learning as a small evidence card:

Roles

owner, worker, reviewer, or specialist responsibilities

Message Contract

artifact, request, response, and handoff state

Coordination

routing, task split, conflict resolution, and final owner

Failure Check

duplicated work, lost context, no accountable owner, or message loop

Eval Action

compare multi-agent result against single-agent baseline

Pass Check

You pass this chapter when a 2 to 3 Agent demo has traceable inputs, outputs, handoffs, final ownership, and a clear reason why it beats a single Agent.

Check reasoning and explanation

1. A passing answer describes the agent loop: goal, plan, tool call, observation, memory or state update, and stop condition.
2. The evidence should include a trace that another developer can inspect, not only the final answer.
3. A good self-check names one safety or reliability control such as tool schemas, permission boundaries, retries, evaluation cases, or a human-review point.