問題文
Inbound documents arrive in three known types: invoices, purchase orders, and shipping notices. Each type has its own extraction logic that is already implemented and tested. The only judgment needed is deciding which type a given document is. Which structure fits?
選択肢
- A workflow that runs all three extraction paths on every document and keeps whichever produced output.
- An agent per document type, each deciding whether the document belongs to it.
- A workflow where a single classification call determines the type and the code then invokes the matching extraction path, so the model makes one decision and the routing stays in code.
- An agent that reads each document and decides at each turn which extraction tool to call next, retrying until it is satisfied.