Workflow Monitoring
Build observability systems to track, analyze, and optimize your agentic AI workflows in production
Your Progress
0 / 5 completedChoosing the Right Metrics
Not all metrics are created equal. The best monitoring systems focus on a small set of actionable metrics that directly inform decisions and drive improvements.
The RED Method for Services
These three metrics provide a solid foundation for monitoring any request-driven system.
Four Categories of Workflow Metrics
Organize your metrics into categories to ensure comprehensive coverage without overwhelming your team:
Speed, responsiveness, throughput—how fast things happen
Success rates, errors, retries—how often things work correctly
Cost, completion, satisfaction—business value delivered
CPU, memory, tokens—infrastructure and cost optimization
Interactive: Metric Explorer
Explore 12 essential workflow metrics organized by category. Select a category to see relevant metrics:
Latency (P50, P95, P99)
PerformanceTime to complete workflow operations at different percentiles
Throughput
PerformanceNumber of workflows processed per unit time
Time to First Token
PerformanceLatency until first LLM response token arrives
Start with 5-7 core metrics that cover the RED method plus your critical business outcomes. You can always add more later, but too many metrics early on leads to alert fatigue and unclear priorities.