Observabilidade

Explains why tracing an agent requires capturing full loops, tool calls, and state, not just requests - the fields and spans that make production agents debuggable.

A first trace showing one agent loop's tool calls and model responses, with minimal Python sketches you can instrument before choosing a platform.

A recipe for instrumenting a LangGraph agent with LangSmith's node-by-node tracing so you can inspect tool calls, state, and branch choices per run.

A recipe for framework-agnostic tracing and self-hosted observability with Langfuse, using OpenTelemetry-friendly instrumentation across agent stacks.

Ten practices for tracing, alerting, and reviewing agent behavior in production - capture, redaction, SLOs, evals, and on-call reconstruction.