AI agent glossary
Tracing and observability
Recording every model call, tool call, input, and output an agent makes, so you can see exactly what happened and why when things go wrong.
An agent's final answer tells you little about how it got there. A trace captures the full path: each prompt sent, each response, each tool called with its arguments and result, along with timing, token counts, and errors.
Observability is the broader practice of collecting and exploring this data. It lets you debug a single failed run step by step, spot patterns across many runs, track cost and latency, and turn real failures into new eval cases.
Set up tracing early, not after something breaks in production. Many agent SDKs and observability platforms support it, and even writing structured logs of each step to a file is far better than nothing.
A user reports a wrong answer. The trace shows the search tool returned an outdated page on step three and the agent never checked a second source, which points straight at the fix.
Learn more in the course
Related terms
Go from definitions to doing
Learn to use and build AI agents in 65 practical lessons.