Making GenAI Observable with OpenTelemetry

GenAI applications don't fail the same way traditional software does. A 200 response means the model answered—it says nothing about whether that answer was useful.

Whitney Lee and Austin Parker, Director of Open Source at Honeycomb, explore how OpenTelemetry helps teams observe GenAI-powered features and coding workflows. They cover why user feedback belongs alongside performance data, why LLMs already understand OTel, and how trace data can help coding agents see what they'd otherwise miss.

OpenTelemetry is a CNCF Graduated project.

Prefer the full ⚡ Enlightning episode?
OpenTelemetry and Generative AI →    • ⚡️ Enlightning - OpenTelemetry and Generat...  

Watch all ⚡ Enlightning episodes →    • ⚡️ Enlightning  

Watch all 🌩️ Thunder episodes →    • 🌩️ Thunder  

Subscribe to Whitney's YouTube channel →    / @wiggitywhitney  

#OpenTelemetry #GenAI #Observability #CNCF #AIAgents #LLM

▬▬▬▬▬▬ ⚡️ Related Links ⚡️ ▬▬▬▬▬▬
🔗 OpenTelemetry → https://opentelemetry.io
🔗 OpenTelemetry GenAI Semantic Conventions → https://opentelemetry.io/docs/specs/s...
🔗 Honeycomb → https://honeycomb.io
🔗 Observability 2.0 Thunder episode →    • Observability 2.0 with OpenTelemetry: Make...  
🔗 CNCF Slack → https://slack.cncf.io

// CHAPTERS //
00:00 - Intro: The GenAI Observability Problem
00:54 - What is OpenTelemetry (OTel)?
02:29 - The 3 Angles of GenAI
03:34 - Angle 1: Training LLMs
04:30 - Angle 2 & 3: Building Features & Workflows
05:55 - How OTel Helps with GenAI-Powered Features
09:10 - Using GenAI in Your Coding Workflow
11:15 - The Challenge of "Architectural Blindness"
13:42 - The Bleeding Edge: What's Next?
15:28 - Final Takeaways