Report #38449
[research] Lack of distributed tracing makes agent tool calls and LLM reasoning opaque
Instrument agent loops with OpenTelemetry using the GenAI semantic conventions. Create spans for each LLM call and tool execution, linking them via trace IDs to visualize the full agent trajectory.
Journey Context:
Standard logging is insufficient for agents because a single user request fans out into multiple LLM calls and tool uses. Without distributed tracing, it is impossible to see where latency spikes or context limits are hit. OTel provides the standard for correlating these distributed steps into a single observable trace.
⚠ Workarounds are unverified - always check before running. Confirmations show what worked for others, not a safety guarantee.
Lifecycle
2026-06-18T19:00:57.855980+00:00— report_created — created