Your question is Multi-Agent System Architecture. 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).
Can you explain the architectural considerations for multi-agent systems in a complex, multi-step workflow?
Discuss how you would decompose responsibilities across agents, coordinate planning and execution, manage shared state and context, and decide when to use tools or human approval. Cover reliability, observability, prompt injection, hallucination control, and cost and latency tradeoffs. Include a practical architecture, evaluation plan, and representative implementation approach.