Multi-Agent Coordination

Problem Statement

Multiple specialized agents collaborate: one classifies, one analyzes, one decides. How do they coordinate? What if one fails? How do they handle conflicting results?

Solution Approach

Delegation Pattern:

User Query
  ↓
Agent A (classifier)
  "This is a fraud case"
  ↓ delegates to
Agent B (fraud analyzer)
  "Risk score: 0.85"
  ↓ delegates to
Agent C (decision maker)
  "Escalate to human"
  ↓
Result

Message Passing:

class AgentChain:
    def run(self, task):
        # A classifies
        classification = self.agent_a.run(task)

        # B analyzes (depends on A's result)
        analysis = self.agent_b.run({
            "task": task,
            "classification": classification
        })

        # C decides (depends on A + B)
        decision = self.agent_c.run({
            "task": task,
            "classification": classification,
            "analysis": analysis
        })

        return decision

Handling Disagreement:

If agents disagree (A says fraud, B says legitimate):
1. Check confidence scores
2. If tie: escalate to human
3. If one confident: use confident agent
4. Retry with different parameters

When to Use

• Use when problem requires multiple specialists
• Use explicit delegation (not implicit; easier to debug)
• Use message passing (stateless communication)

References

• Distributed Tracing
• Recovery Strategies