Development

Tracing agent and model runs

• Span Attributes That Let You Slice by Model, Tenant and Outcome
• Structuring Logs and Traces for AI Agents: What to Record
• Tracing the Query Path: Spans for Retrieval, Reranking, Calls
• A Trace Attribute for Remaining Time at Each Tool Call
• Building a Trace Viewer That Answers the First Question
• Building the Agent Status Page Your Support Team Reads
• Choosing a Tracing Backend for Agent Runs
• Finding the Failing Step in a Trace Without Reading Logs
• Instrumenting Agent Workflows: What to Log, Trace, and Alert On
• Per-Tenant Error Rates from Span Attributes
• Reading a Prompt Injection in Traces: Fields and Triage Order
• Reading a Retry Storm in Traces by Tool, Tenant and Prompt
• Reading a Spend Spike from Traces: Model, Customer or Retry Loop
• Reading a Validation Failure in Traces: Which Field, Which Model
• Reading Traces: Model Fault or Upstream Fault
• Sampling High-Volume Agent Traces Without Missing Errors
• Span Events vs Child Spans for Retrieval and Model Calls
• Spans for a Multi-Step Agent Run: Where Each One Starts and Ends
• Telling a Policy Denial From a Model Timeout in Traces
• Telling a Truncated Completion from a Refused One: Trace Signals
• Trace Query Primitives That Answer Incident Questions
• Tracing a Read from Edge Cache to Replica and Back
• Tracing a Slow Request: Spans from Edge to Model and Back
• Turning Span Attributes into Tenant Dashboards and Alerts
• What Trace Attributes a Prompt Version Needs Beside It
• When the Agent Loops: Breaking Repetition from Traces
• Where Injection Attempts Land in Agent Traces