Whenever you're ready, walk me through how you'd approach this.
How would you implement a multi-agent system for an AI task at Bain &?
Clarify the task, users, success criteria, data access, and operational constraints before proposing an architecture. Explain how agents would be decomposed, coordinated, evaluated, deployed, and monitored, including failure handling and training-serving consistency.