← Evals
Core

Agent Traces in Production: Spans, Metrics and Turning Failures into Evals

An agent trace records one run as nested spans: the agent invocation, each model call and each tool call, with timing, tokens, errors and identifiers. Learn what a trace shows that a log line cannot, which product metrics come straight from traces, how the emerging OpenTelemetry conventions name the spans, and how to reconcile tracing with zero-retention customers.

CONCEPT / EVALS
FREE ACCOUNT / CONCEPT

Read this concept with a free account

Evals

Read an agent run as a tree of model and tool calls, then turn its failures into test cases.

  1. 01

    Understand the mechanism

    Read the full explanation behind this product decision.

  2. 02

    Use it in an interview

    Follow the related practice questions and put the idea into an answer.

Your free account includes this concept. No card needed.

Read a free concept in this topic
RELATED CONCEPTS
PRACTISE THIS CONCEPT