WHY USE MORE THAN ONE AGENT

WHEN NOT TO USE MULTIPLE AGENTS A responsible engineer knows when to say no. If a task is short and straightforward, such as answering a customer's question about store hours, a single agent is better. It is cheaper because there are fewer calls. It is faster because there are no handoffs. It is easier to debug because there is one transcript to read. A helpful rule of thumb is to start with one agent and measure how well it does. If it is failing in a specific way that splitting would fix, such as a cluttered context or conflicting instructions, then consider splitting. Do not split first and hope. A good harness engineer asks three questions before adding a second agent. First, is the task genuinely made of distinct parts that need different skills, tools, or mindsets? Second, is the single agent suffering from context clutter or from instructions that fight each other? Third, is the extra cost and complexity worth the improvement I can actually measure? If the honest answer to all three is not a clear yes, stay with one agent.