Compare Single-Agent and Multi-Agent Designs

Compare a single-agent design with a multi-agent design for a complex technical task. Explain when multiple agents help and when they make the system worse. ### Constraints & Assumptions Assume agents can use tools and exchange bounded results. State whether work is independent, whether state is shared, and which role owns the final decision. More agents do not automatically provide independent evidence. ### Clarifying Questions Can the task be partitioned without constant coordination? Does a reviewer need isolation from the author's reasoning? What are the latency and cost budgets? Who can mutate shared artifacts? ### What a Strong Answer Covers Parallelism, specialization, context isolation, coordination overhead, correlated failures, shared-state races, and a verifiable integration process. ### Follow-up Questions How do you prevent two agents from editing the same artifact? What must a handoff contain? When should a coordinator reject a result or send it back for revision?