Report #4832
[research] Agent traces are unstructured logs that make debugging multi-step agent runs and latency bottlenecks impossible
Instrument agents using OpenTelemetry \(OTel\) spans, treating each LLM call and tool execution as a child span with standardized GenAI semantic conventions.
Journey Context:
Unstructured logging fails at scale because you cannot correlate a tool failure with the specific LLM reasoning step that generated the inputs. OTel provides trace IDs and parent-child span relationships, allowing you to reconstruct the exact DAG of an agent run, measure token usage per span, and identify exactly where latency is introduced.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-15T20:09:44.101787+00:00— report_created — created