CONTEXT ENGINEERING FOR SELF-HEALING AI SRE
summary
Komodor analyzes nearly 1.8M incidents and reduces Kubernetes chaos by moving from brittle deterministic runbooks to an AI agentic system named Claudia agent that runs dozens of subagents, uses Cloud Code integrations and per-customer MD blueprints for CRDs, and evaluates changes with shadow-agent evaluation. Production stats: 32K investigations, 12 agents, ~77s median latency, and a 99.7% pinpoint rate for identifying the correct remediation, with many fixes applied autonomously.
AI-written summary · about
chapters (10)
- 0:00Introduction and speaker
- 1:35Commodore and Kubernetes management
- 9:25Challenges with deterministic runbooks
- 15:53Same issue different root causes
- 23:03Why deterministic runbooks failed
- 25:55Switching to AI with Claudia
- 31:43Evaluation and shadow agents
- 34:11Claudia’s performance metrics
- 38:03Build vs buy AI SRE agents
- 41:39Final thoughts and conclusion
related talks
40mWhen Incidents Fix Themselves: AI SRE in action
44mAgentic Observability — The Path to AI Co-SRE
17mFighting AI with AI
27mAI-Powered Root Cause Analysis at Scale: From Theory To Production...
15mIntelligent Incident Management in DevOps
30mLive Panel with
22mHidden Signals in K8s Clusters
38mAI as a Platform Engineer: Explaining Kubernetes Failures, Not Just Detecting Them