Multi-agent when not to
When is a multi-agent architecture overkill compared to a single LLM with tools?
Answers use simple, clear English.
Quick interview answer
If the task is single-turn Q&A, simple RAG, or one/two tool calls, multi-agent adds latency, cost, and failure modes without gain. Use multi-agent when roles truly specialize (research vs code vs review) and you have evals showing improvement. Prefer a single agent + tools + clear policies first.
Detailed answer
If the task is single-turn Q&A, simple RAG, or one/two tool calls, multi-agent adds latency, cost, and failure modes without gain. Use multi-agent when roles truly specialize (research vs code vs review) and you have evals showing improvement. Prefer a single agent + tools + clear policies first.
Real example & use case
FAQ bot: single RAG agent; codebase migration: planner + coder + reviewer agents.
Pros & cons
Pros: simplicity wins early. Cons: complex workflows may need roles later.