When the Codebase Starts Instrumenting Itself
summary
Builds an AI agent called Spiny-backed Orb-weaver that reads a Weaver registry, instruments JavaScript code with OpenTelemetry spans, updates the registry and opens a PR branch with companion markdown explaining choices. It uses the OpenTelemetry Weaver CLI, an Entropic API key, an init file and OpenTelemetry API; runs a deterministic orchestrator with retries and a live OTLP validator every five files that runs the test suite. A 30-file run takes ~40 minutes and reports 30-file repo: 12 committed, 17 skipped, 1 failed with a $67 ceiling and $4.36 actual cost.
AI-written summary · about
chapters (14)
- 0:00Self-instrumenting codebases intro
- 1:28Manual instrumentation pain
- 3:11Why code telemetry matters
- 4:35Spiny-backed Orb-weaver
- 5:58OpenTelemetry Weaver registry
- 8:07How Spiny-backed Orb-weaver works
- 11:17Instrumentation agent architecture
- 13:44Safe instrumentation rules
- 16:08Instrumentation score references
- 16:50Demo: uninstrumented code
- 17:58Demo: instrumented code
- 19:45PR-ready instrumentation report
- 22:37Schema extensions
- 23:38Q&A
related talks
23mObservability by Design: Leveraging OpenTelemetry Weaver To Take Control of Your Observability
18mBuilding an APM with OpenTelemetry and OpenSource
55mCoding with the lights ON - OpenTelemetry from code to prod
45mModern Observability: Making OpenTelemetry work for you
47mExploring a complex codebase with AI
51mCoding with the lights ON - OpenTelemetry from code to prod
31mDeep Dive To AI Agent Observability
32mFrom Toil To Triumph: Harnessing Agentic AI To Streamline Infrastructure as Code