trace

reasoning trace explanation traces

Backlinks

blog post - the three pillars of AI observability - 2025-11

traces: Reconstruct the full decision path for a request, across model calls, tools, retrieval, and control flow

see in context

Users also don't want to jump between three disjointed experiences metrics, logs, and traces to create a full picture of user behavior, so in AI, these have collapsed into one thing: tracing
...
AI observability metrics, logs, and traces collapse into traces

see in context

The core primitive of evaluation is scoring. Scoring allows you to look at an entire agent interaction (trace) or turn (span) and quantify it. Usually, this means producing a number (for example, how factually grounded is the answer?), but it can also be categorical (for example, what type of error is this?). The best teams use online evals to help them discover what to test in dev and CI.
...
using online evals to help discover what to test in dev and CI

see in context
free recall - Open AI harness blog post - 2026-02

The second principle was strive for dim - legibility -- higher for the agent... AKA agent legibility. This means giving the agent everything it needs to get the job done: context right in the repo specs right alongside the code, all the logs, traces, etc. it also means keeping the code base structurally legible. Obvious folders, strict and clear naming conventions, etc.

see in context
metrics, logs, and traces

link not tracked link not tracked and traces

The workhorses of standard application observability

see in context
trace span

^ trace span

see in context
traces should provide reproducible context, not just spans

traces should provide link not tracked, not just spans

see in context