Your question is Multi-Agent Reasoning Design. Take a moment with it on the right.
Talk me through your thinking if you like. When you're confident, submit your answer and I'll grade it like a real screen (7/10 or better passes).
How do you design and implement multi-agent systems for complex, multi-step reasoning tasks?
Discuss how you would decompose work among specialized agents, coordinate their interactions, select tools and models, and define termination and recovery behavior. Include implementation details, evaluation before deployment, cost and latency controls, hallucination prevention, and defenses against prompt injection. Provide pseudocode or Python using a modern LLM SDK.