問題文
A multi-agent research system uses a coordinator with four subagents: web search, document analysis, synthesis, and report generation. Currently each subagent sends its output directly to the next subagent. The team wants to know which topics failed and why, and to apply consistent error handling. Which change best addresses this?
選択肢
- Add a retry wrapper around each subagent call so transient failures do not surface at all.
- Route all inter-subagent communication through the coordinator so it observes every result and handles every failure in one place.
- Merge the four subagents into a single agent so there are no inter-agent handoffs to observe.
- Have each subagent write its own log file and add error handling code inside each subagent.