Agent orchestration coordinates agents through a workflow, but successful handoffs cannot be assumed. Each step should remain visible, with enough context for a human to review decisions and clear ownership when something goes wrong. section 6.7
Watch for risks that arise as agents interact:
- Miscoordination: agents share a goal but take mismatched actions, such as a broken handoff that leaves a claim unanswered. section 6.7
- Conflict or collusion: agents may work against each other or cooperate in ways that produce harmful outcomes. section 6.7
- False independence: agents may agree because they share a model, retrieval index, prompt template, or business objective—not because their checks are independent. section 6.7
A handoff that never arrives
A customer-service chatbot passes a claim to another agent, marks the handoff complete, and closes the conversation. A broken integration prevents the second agent from receiving it, leaving the claimant without help. section 6.7


















