A team builds a customer service automation system where one LLM agent handles the entire conversation: understanding the customer's issue, querying the database, checking policies, and drafting a response. The system works for simple cases but fails on complex multi-step disputes that involve billing, technical support, and legal compliance simultaneously. A colleague proposes splitting into multiple specialized agents. What is the fundamental architectural motivation for multi-agent systems over single-agent systems?