SRE'ing the SRE AI Agents
summary
An operational prototype uses LLMs plus a vector DB for RAG, Prometheus/Elasticsearch/tracing for realtime context, and an MCP layer that both reads metrics and invokes prebuilt scripts so the agent can open PRs, declare incidents and remediate. It runs in POC mode with metrics: target 95% accuracy SLO over 30 days, rollback rate <2% and example MTR ~50 minutes; accuracy shows near‑100% but occasional drops trigger human rollbacks. Key controls: permission boundaries, hierarchical RBAC and MCP remediation with cryptographic attestations, retracting autonomy if accuracy <90% for three days.
AI-written summary · about
chapters (10)
- 0:00Introduction and agenda
- 3:34Why we need AI agents
- 9:30Internal mechanics of agents
- 14:27Autonomy and impact
- 16:12Permission management
- 17:15Security and access control
- 19:30Audit trails and compliance
- 22:15Observability and metrics
- 26:59Lessons and takeaways
- 34:07Q&A session
related talks
40mWhen Incidents Fix Themselves: AI SRE in action
44mFrom 4 Hours to 8 Minutes with AI Agents That Transform SRE...
36mModernizing Incident Response with LLMs, RAG, and the MCP
15mWho Owns Your SRE Stack? The Rise of AI Agents
34mAutomating SRE Operations with Multi-Agent AI
25mLLM Agents for Site Reliability
27mAI Teammates: How will they impact SRE Teams?
25mLLM Copilots for SRE: Incident Response & Auto-Remediation