Observability as a foundation for AI-enabled platforms
summary
OpenTelemetry graduates as the recommended foundation, with ~49% of orgs running it in production and annual growth around 35%, and the advice is to auto-instrument every layer using the OpenTelemetry operator so traces, metrics, logs and profiling are standardized. Use the GNAI normalizer or Arconia to align different semantic conventions, export agent telemetry, and expect sparse reasoning signals (demo showed 290 spans with only 15 containing explicit reasoning). Platform teams build observability as a product to enable safe, incrementally autonomous agents.
AI-written summary · about
chapters (14)
- 0:00Introduction
- 1:15OpenTelemetry for AI platforms
- 3:20Platform engineering in AI era
- 6:56Missing observability in AI talks
- 9:03Observability for platform engineering
- 10:28Observing AI workloads
- 13:17Tracing agent interactions
- 14:19Trust in AI agents
- 17:46Agent autonomy levels
- 19:51Agentic SDLC feedback
- 21:36Telemetry for coding agents
- 22:18Production agents on Kubernetes
- 23:42Key takeaways
- 25:45Closing
related talks
27mObservability as a foundation for AI-enabled platforms
07mKeynote: OpenTelemetry And The Future of Open Source Observability
31mDeep Dive To AI Agent Observability
20mGenAIOps - OpenSearch for AI & GenAI Platform Observability
34mObservability 2.0: Shifting Left From Reactive Mon...
04mOpenTelemetry Celebrates Graduation and the Next Era of Agentic...
28mDO WE STILL NEED TO "OBSERVE" THE FUTURE OF AI & OBSERVABILITY
09mSetting up Intelligent Observability