A retail company wants an automated returns system where intake, fraud screening, refund approval, and customer notification happen in sequence, with a human approving refunds above a threshold. How should the goal be decomposed?
Select an option, then click Submit answer.
- ○ Separate agents for intake, fraud screening, and refund calculation, orchestrated sequentially with a human-in-the-loop approval step for high-value refunds
- ○ One general-purpose agent that handles every return end to end with no checkpoints
- ○ A single rules engine with no agents, because returns never need language understanding
- ○ Parallel agents that all approve refunds independently and notify the customer multiple times