Agent Observability Playground

3 scenarios exploring observability for AI agents in production. Toggle logging, tracing, metrics, and alerting to see how they improve your ability to debug and fix production issues.

Choose a Scenario

3 scenarios available

Production Incident

An AI agent in production is failing silently — users report wrong answers but no errors appear in your dashboard. Each observability component you enable gives you more visibility into what went wrong. Watch how your debugging ability improves as you build up the observability stack.

Incident

Silent failures — wrong answers, no errors

Agent

RAG-based support agent — GPT-4o

Impact

~15% of queries returning hallucinated answers

Token Usage

0/ 1,050 (0%)
Logging
Tracing
Metrics
Alerts
Costs
Replay

Context Components

0/6 active

0/100

Broken

Completely Blind

No components

Engineer

Our support agent has been returning incorrect policy information to customers since last night. No errors in the logs, no alerts fired. Multiple customers have complained. I need to figure out what broke and fix it before more customers get wrong answers.

AI Agent

Completely Blind

Issues (5)

  • ×Zero visibility into agent behavior
  • ×Cannot determine root cause
  • ×Cannot scope the blast radius
  • ×Cannot verify if a fix worked
  • ×Reduced to guessing and restarting