Agent Metrics
Agent metrics are quantitative measures of agent performance. They enable comparison across versions, alerting on degradation, and data-driven optimisation.
6 min•By Priygop Team•Updated 2026
Key Agent Metrics
- Task success rate: percentage of tasks that complete successfully without escalation
- Steps per task: average number of steps taken to complete a task — lower is generally better
- Tool selection accuracy: percentage of tool calls that were the optimal tool for the situation
- Escalation rate: percentage of tasks that required human escalation
- Retry rate: percentage of tool calls that had to be retried
- Latency: average time from task submission to completion
- Cost per task: average token cost to complete one task
- Error rate: number of unrecoverable errors per 100 tasks
Key Takeaways
- Agent metrics are quantitative measures of agent performance.
- Task success rate: percentage of tasks that complete successfully without escalation
- Steps per task: average number of steps taken to complete a task — lower is generally better
- Tool selection accuracy: percentage of tool calls that were the optimal tool for the situation