An agent trace and an activity log solve different problems. A trace is a microscope — it shows you the internal detail of one specific run. An activity log is a telescope — it shows you the pattern of what your agent has been doing across many runs over time.
What Each One Is For
An agent trace captures the complete internal story of a single execution: every reasoning step, every tool call, every decision point from start to finish. You use it when something went wrong in a specific run and you need to understand exactly what happened. It’s detailed, verbose, and focused on one event.
An activity log is a higher-level record of agent behavior over time. In the Claude Cowork system, this is the wp_agent_logs table — it records what skill ran, when, its status (success, error, partial), how long it took, and a brief summary of what it did. You use it to answer questions like: “Did the morning report run today?” or “How many articles has the TAM writer published this week?” or “Which agent has been failing most often?”
How They Work Together
The activity log tells you something is wrong. The trace tells you what the something is. If your activity log shows that a particular agent has had three errors in the past week, you pull the trace from one of those error runs to understand the cause. The log surfaces the pattern; the trace explains the instance. Neither one alone gives you the full picture.
Think of it like a hospital’s patient records system. The activity log is like a patient’s visit history — dates, outcomes, general status. The trace is like the doctor’s detailed notes from a specific visit — what was observed, what was tested, what decisions were made. You read the visit history to spot a pattern; you read the detailed notes to understand a specific event.
What This Means for Educators
For campus operators, setting up both layers of logging from the beginning — an activity log for ongoing oversight and detailed traces for diagnostic use — is the operational foundation that makes agent-powered systems manageable at scale.
The Simple Rule
Check your activity log weekly for patterns. Pull traces when you find something worth investigating. Don’t try to read every trace — that’s what the activity log is for.