How do you examine what an agent did at each step?
Why C is the answer
MLflow tracing
MLflow tracing records each step, tool call and result, so you can see what an agent actually did.
In detail
- MLflow scoring and tracing evaluate agents.
- A trace records each step, tool call and result.
- Attach traces to evaluation results.
Use case
An illustrative scenario, not a real incident.
An agent gives a wrong answer. An MLflow trace shows each step, and that the second tool call returned nothing, so the team knows where to look.
See it in different ways
flowchart LR A["Evaluate on a test set"] --> B["Deploy"] B --> C["Monitor live traffic"] C --> D["Add failures to the test set"] D --> A
sequenceDiagram participant U as Caller participant G as AI Gateway participant E as Endpoint participant T as Inference tables U->>G: Request G->>G: Apply rate limits G->>E: Forward E-->>U: Response E->>T: Record the request and response
flowchart LR S0["User input"] --> S1["Model call"] --> S2["Tool call"] --> S3["Model call"] --> S4["Final answer"]
Key point: Traces show what happened at each step.
Why the other options are wrong
- A. A Delta constraint
- A constraint checks table data.
- B. The model card
- A model card describes a model, not one run.
- D. The row count of the inference table
- A row count shows volume, not the steps.