Observability (agent tracing)
Recording every step, tool call, token count and latency of an agent run so it can be inspected.
Because agents take many non-deterministic steps, debugging requires a trace: the full sequence of prompts, model outputs, tool inputs/outputs and timings. Tracing is the most-cited control in LangChain's surveys — 89% of organisations have some observability and 62% trace individual steps (94% and 71.5% among teams in production). Tools include LangSmith, Langfuse, Arize and general APM systems.