Hi,
I’m building agentic workflows, but when something fails Orchestrator just says “Job failed” with no visibility into what the agent decided, why it took that path, or what prompt/response caused the failure.
Please suggest how can i do debugging and RCA(Root Cause Analysis) for agent behaviour in production?
Orchestrator does not currently provide visibility into agent reasoning, prompt/response history, or decision paths. It only reports job status, logs, and exceptions.
For debugging and RCA of agentic workflows, observability must be built into the process by explicitly logging and persisting agent prompts, responses, decisions, and versions (e.g., via structured logs, queues, or external storage).
It might help you
Cheers
Hi @Aniket_Baikar,
Orchestrator alone is not enough for RCA in agentic workflows. You need to add observability in the agent. Log prompts, inputs, model and prompt versions, tool calls, intermediate decisions, and outputs with a correlation id. Wrap tool calls in Try Catch and log before and after each step. Use a debug mode via config or asset to increase logging in production. RCA then comes from reviewing these logs, not just the Job failed status.
Hope this works
Thanks for explanation
1 Like
This topic was automatically closed 3 days after the last reply. New replies are no longer allowed.