Glossary · 6 · Quality, risk and governance
Agent trace
Also known as: Agent run log, Tool-call log, Agent observability
German: Agenten-Trace
An agent trace is the structured, time-stamped record of one AI agent run: the task and its origin, the model and configuration, each reasoning step the platform exposes, every tool call with its parameters and result, the approvals requested and given, and the final output — linked so that the run can be reconstructed afterwards.
- Intermediate
- Developers
- Technical writers
- Technical project managers
In one sentence
An agent trace records one AI agent run step by step — task, model, tool calls, approvals, output — so it can be reconstructed afterwards.
Example
The trace of an agent that deleted calendar entries shows the e-mail it read, the instruction hidden in it, the three tool calls that followed and that no approval was requested.
Explained in context
This term is a station of a mini course
- When an AI agent crosses the line: detect, contain and restore operationsStation: Agent trace and tool-call log · Preserve the evidenceStart the course
Context cards connect this term with others to answer one question. Also in British English and German.
- Detecting an AI agent that crosses its boundaryHow do you notice that an AI agent has done something it wasn’t allowed to do — before a customer, an operator or the month-end report does?UKDeutsch
- When an AI agent crosses the line: detect, contain and restore operationsWhat has to be in place when an AI agent does something it isn’t allowed to do — in IT systems, IoT fleets, OT and on machines — and how is it detected, stopped and released again?UKDeutsch
How it applies
- Results don't prove the path: An agent can reach a good result through a forbidden action. Only the trace shows which tools were called and whether a boundary was crossed.
- What to record: Task and who started it, model and version, system prompt version, retrieved sources, tool calls with parameters, refusals, approvals, errors, output. Mask personal data and secrets.
- Standards: Agent platforms offer their own traces; OpenTelemetry is developing semantic conventions for generative AI that make traces portable across tools (status “Development” as of October 2026).
- Evidence: In an incident the trace is the main evidence for triage, for reports to authorities and for the review. Keep it tamper-evident and for a defined period; for high-risk AI systems see record-keeping and logs.
Why it matters on your learning path
- Technical writers: Traces show how an agent really used your content — a source for improving it.
- Technical marketers: Traceability is a concrete trust argument; show an example trace.
- Technical project managers: Set retention, access and review routines for traces.
- Developers: Trace every tool call with a run id that links to the systems it touched.