앤트로픽·오픈AI·마이크로소프트 모두 같은 말…”에이전트, 꼭 필요할 때만 써라”

Leading AI companies—Anthropic, OpenAI, and Microsoft—are issuing a unified warning: most developers over-engineer with multi-agent systems when simpler solutions suffice. They recommend optimizing single LLM calls with better retrieval design and prompt engineering before jumping to agent orchestration, mirroring the microservices complexity lesson from the previous decade. The consensus is clear: modularity comes at a cost, and unnecessary distribution introduces observability, debugging, and maintenance burdens that often outweigh benefits.