Beyond Dashboards: Architecting AI Agents for Autonomous Observability
summary
Claims an RCA pipeline that reduces typical 90 minute investigations to ~30–90 seconds by combining deterministic math with LLMs and RAG over a vector DB. Architecture uses topology-driven BFS on a service map, parallel signal fetchers to collect metrics/logs/traces into an MCP evidence layer, deterministic correlation windows (causal gaps) and then LLM nodes for English interpretation, hypothesis generation and RAG-backed validation. Demo on a 9-service OpenTelemetry e‑commerce app (Prometheus, Jaeger, Loki, Grafana, LangGraph) returns RCAs in 27–37s with example confidences 93%, 83%, 52%.
AI-written summary · about
chapters (8)
- 0:00Introduction
- 2:10What is RCA and why hard
- 6:07Four pillars of autonomous RCA agents
- 9:40Architectural walkthrough of RCA agent
- 16:00Live demo of RCA agent
- 21:17Medium complexity RCA demo
- 24:44Hard complexity RCA demo
- 26:50Key takeaways and conclusion
related talks
32mGenerative AI for Observability: Automating Root Cause Analysis in Modern IT Systems
07mAccelerating RCA with AI
46mLeveraging AI and Automation in DevOps
30mAI Agents for Incident Management
16mAI Agents for Incident Investigation: The Good, The Bad, and The Ugly
23mTrusted Agentic AI:Architecting Systems with Reliable Data at Scale
16mMulti-Agent Intelligence for Next-Gen Insights
32mFrom Three Days to One Minute: Architecture Trade-offs in Production Enterprise AI