You are designing an LLM-powered customer support assistant. It needs to answer common product questions, look up account details, summarize prior tickets, and decide when to hand off to a human. Your team is debating whether to use one general-purpose agent with several tools, or a multi-agent setup with separate agents for triage, retrieval, policy checks, and response drafting.
Compare single-agent and multi-agent architectures for this customer support task. How would you decide between them, and what tradeoffs would matter most in practice?