Agent trace
Also called agent tracing, run trace, LLM tracing
A step-by-step record of one AI agent run: every model call, every tool call and its result, under one run ID.
An AI agent works in a loop: it calls the model, the model asks for a tool, the tool returns something, and the model decides what to do next, many times over. When the run goes wrong, the question is always what the agent saw at the moment it went wrong.
A trace answers that. Every step of one run is recorded under the same ID, in order, so you can read it like a transcript: the exact prompt, the tool call it chose, the result that came back. The cause usually turns up earlier than you'd expect, in a stale file, an empty search, or an error the agent took for success.
For example
An agent edits the wrong config file. Its trace shows that at step 4 a search came back empty, and the agent guessed a path from there.
Put it to use
Where you'll meet agent trace in ChatGPT, Gemini and Claude, how to use it there, and the mix-up most people make.
Daily Bytes Plus is in the Android app.