A rare multi-agent failure occurred once in production. Engineers c…

Question

A rare multi-agent failure occurred once in production. Engineers cannot reproduce it from logs that only store final answers. Which observability capability is most important to add?

Accepted Answer

Agent replay capture with structured logging of reasoning paths and tool I/O (within policy) plus cross-service trace correlation so failures can be reproduced offline — A matches observability components in the architect/evaluate space: structured logging of agent reasoning paths, agent replay capture, and cross-service trace correlation for reproducible debugging. Final-answer-only logs are insufficient for multi-agent forensics. B OS syslog alone lacks agent semantics. C rewriting everything is not observability. D removing correlation IDs makes stitching worse. Capture enough prompt/tool/state detail—safely—to replay the rare path.

Answer

Increase log retention for OS syslog only

Answer

Replace all models with rule engines immediately

Answer

Delete correlation IDs to simplify storage schemas

A rare multi-agent failure occurred once in production. Engineers cannot reproduce it from logs that only store final answers. Which observability capability is most important to add?

Short Explanation

Full Explanation