AI as a Platform Engineer: Explaining Kubernetes Failures, Not Just Detecting Them
Talkat Open Source Summit India 2026 · on Tuesday, 2026-06-16· at 14:50· Lotus Ballroom 3 (Level 3)·
summary
Comcast describes an AI explanation layer that cuts Kubernetes deployment troubleshooting from 30–45 minutes to about 5 by combining declarative JSON rules (failure domains, ownership, signal lists) with LLM summarization at temperature 0, PII scrubbing, and structured four-field output (responsible party, classification, root cause, remediation) written back into the custom resource status. It runs in two modes (automatic on failures and operator-invoked), handles multi-region private clouds and ~400 clusters with 30+ add‑ons, and averages 8k–16k tokens per request.
AI-written summary · about
chapters (9)
- 0:00Platform engineering and AI
- 3:11Why platforms exist
- 8:50AI explains Kubernetes failures
- 12:21Building AI explanation layers
- 21:25AI for day two operations
- 23:52AI for platform onboarding
- 28:50Evaluating AI agent performance
- 31:50Platform vs application failures
- 37:50Q&A
related talks
22mAI in Kubernetes The Promise
17mSponsored keynote: Can AI handle the complexity of Kubernetes and cloud-native environments?
30mAI Agents for Incident Management
28mAI for Everything After Code
26mAI that is gold does not glitter - using intelligence in Modern Infra
09mAI and Chaos Engr: Smarter Failure Testing for Resilient Systems
12mAI-Driven Platform Engineering
38mAI in Kubernetes: How to Get Started?