Asimov's Zeroth Law of Robotics: Observability for AI
Talkat KubeCon + CloudNativeCon Europe 2025 · on Wednesday, 2025-04-02· at 16:15· Level 1 | Hall Entrance N10 | Room E·
summary
Argues a zeroth law: AI must be observable so other laws can be verified, and shows a concrete instrumentation stack: OpenTelemetry, OpenLit, Prometheus, Tempo, Loki and Grafana Cloud capturing tokens, temperature, model/version, cost, latency, retries, traces and logs. Demonstrates a Flask Python two‑player D&D demo instrumented end to end, then runs systematic tests and chaos/load scenarios with K6 load tests to detect hallucination, bias, rate‑limit faults and model drift and to compare model versions in CI.
AI-written summary · about
chapters (13)
- 0:00The laws of robotics
- 1:18Robot logic failures
- 4:12The zeroth law of robotics
- 5:47Observing AI in practice
- 8:39Telemetry signals for AI
- 11:00Instrumenting with OpenLIT
- 12:10Demo: D&D AI assistant
- 16:32Visualizing AI performance
- 18:21Testing AI for reliability
- 20:03Systematic testing with k6
- 24:13Advanced testing strategies
- 26:21Future AI testing approaches
- 28:30Lessons from robot logic
related talks
25mWhen Your "New Teammate" Hallucinates: Observability as the Safety Net for AI Code
28mDO WE STILL NEED TO "OBSERVE" THE FUTURE OF AI & OBSERVABILITY
16mWhat Did the AI Just Do?: Making LLMs Observable with Open Telemetry
31mBuilding Trustworthy AI: Open Source Monitoring in Action
25mAI Observability: The Missing Runtime Control Layer for API-Driven Systems
40mKeep AI on Track: Guardrails + OpenTelemetry Observability
28mWhere are we going with observability?
18mAI Observability: How To Fix Issues With Your ML Model