CONCEPT Cited by 1 source
Agent observability trace tree¶
Definition¶
Agent observability trace tree extends distributed-tracing concepts (spans, hierarchical parent-child relationships) to LLM agent reasoning loops. Each orchestrator decision, tool invocation, latency breakdown, and token cost is captured as a span in a hierarchical trace tree. Engineers can drill from a top-level orchestrator span down through each reasoning iteration to pinpoint root causes.
The analogy: debugging a 40-step research task works like debugging a distributed microservice call — except the "services" are LLM reasoning steps and the "RPCs" are tool calls. (Source: sources/2026-06-18-atlassian-long-horizon-reasoning-engine)
Why it matters¶
Traditional request-scoped logging was never designed for reasoning loops spanning dozens of iterations and tool calls. Without trajectory-level observability, it's impossible to diagnose: - Wrong tool selection at step N - Silent failures masked by the model's recovery attempts - Retries that burned iteration/token budget - Context window pressure building across iterations
Seen in¶
- sources/2026-06-18-atlassian-long-horizon-reasoning-engine — Atlassian instruments Long Horizon with structured LLM tracing as hierarchical trace trees.