Sentry's Span Hierarchy Exposed a Silent Retry in My 5-Agent Pipeline. One Agent Took 22.6s, the Others Took 5.

This is a debugging write-up, not a tool, but it belongs in the reading list for anyone shipping multi-agent pipelines. The author used span hierarchy tracing to find that one agent in a five-agent system was taking 22.6 seconds while siblings took five — not because of a slow model, but because one tool was generating seven times more token output than expected, silently triggering retries. The fix is pragmatic: add pagination to the offending tool and enforce a token budget guard so agents cannot spiral on runaway output. Numbers are concrete: 42% reduction in output, 21% faster end-to-end. The broader lesson is that silent retries in agent pipelines are nearly invisible without span-level tracing, and most teams are not instrumenting at that level yet. If you are running agent pipelines in production without this kind of observability, this post will give you a specific thing to go add today. -> Best for: technical PM or AI engineer running multi-agent systems in production