The port-ops copilot is being scoped, and Meridian's product owner asks: 'Should this be a single conversational assistant, or should we split berth-schedule questions, circular questions, and crane-status questions into three separate agents coordinated by a supervisor?' Under what condition does the three-agent-plus-supervisor design become the better choice over a single agent?

The three-agent-plus-supervisor design becomes the better choice specifically when the underlying question types draw on meaningfully different data sources, tools, or specialized logic - berth schedules being a live system query, circulars being document retrieval, and crane status being another live system - because a single agent would otherwise need to juggle multiple unrelated tool sets and decision logic inside one prompt, increasing the chance of misrouting or tool-selection errors. Claiming the multi-agent design is always better regardless of how distinct the domains are is wrong because added orchestration complexity is not free - if the three question types were actually similar in data source and logic, a single agent would be simpler to build, debug, and maintain, and splitting it up would be unjustified overhead. The claim that Databricks discourages multi-agent designs is false; Multi-Agent Supervisor is an officially supported Agent Bricks pattern specifically for this kind of domain-routing scenario. Choosing purely by token count per request ignores the actual design driver, which is whether the domains genuinely require separate specialized logic - token efficiency is a secondary optimization, not the basis for the single-agent-versus-supervisor decision. This item tests the conditional judgment behind the Agent Bricks decision, not just pattern recognition from a keyword.