Hamel Husain Explains What a Trace Is in LLM Evaluation
HamelHusain · x · 2026-09-19
In his AI Evals FAQ, Hamel Husain defines a trace as the complete record from a user's first query through the final response — every step across all agents, tools, and system components in a session, including multiple user messages, assistant responses, retrieved documents, and intermediate tool calls.
He also notes terminology drift: observability vendors define traces and spans differently (citing Alex Strick van Linschoten's comparison). The piece is part of his AI Evals FAQ series.
More from coding & agent
- OpenHarness Open-Sources a Platform to Run Claude Code and Codex Across All Your Machines — dee_hw · 2026-09-19
- How to use Jev well: give it the decision you make 1,000 times, not your whole project — JensHonack · 2026-09-19
- Greg Kamradt launches agnts.sh: a message board built for AI agents — GregKamradt · 2026-09-19
- EDA team building MCP server for CAD tool asks: how many tools is too many? — Important_Earth6615 · 2026-09-19
- 3 Franka arms stack 9-block tower with Gemini Robotics-ER as VLA brain — Zergylord · 2026-09-19
- Claude Code Max overage pricing: 2 extra days could cost $925+ on top of $200/mo — julianharris · 2026-09-19