Keynote: Tracing the Agent's Mind: Extending OpenTelemetry for Deep...
summary
Production agents make risky tool calls; this work introduces the gen-ai-otel extension that adds a gen_ai.agent. attribute namespace capturing tool-selection confidence, permission scope checks, session context and baseline deviation, and ships zero-sidecar architecture routing through standard OpenTelemetry Collectors to Jaeger/Prometheus. A live demo reconstructs decision chains and cuts decision-debugging time; benchmarks cover 73 tools across 5 risk tiers and an 8,100-chain benchmark. Code, Collector configs and Grafana dashboards are Apache‑2.0 and ready to deploy.
AI-written summary · about
chapters (7)
- 0:00AI agent decision risks
- 2:06Tracking agent decisions challenges
- 5:56OpenTelemetry for agent decisions
- 17:07Benchmarking agent behavior
- 20:19Inverse scaling trade-offs
- 23:38Key takeaways and future needs
- 25:36Open-source implementation tools
related talks
33mWhat Did My Agent Do? Observability and Accountability for AI Agents
31mDeep Dive To AI Agent Observability
30mBuilding Reliable AI Agents: An Open Source Approach To Evaluation and Observability
25mLet Me Be Your OpenTelemetry Champ
27mObservability as a foundation for AI-enabled platforms
38mPractical introduction to OpenTelemetry tracing
30mTutorial: Building an Observable AI Agent With an Open Source Stack
43mPractical Introduction to OpenTelemetry Tracing