Anthropic Reliability Engineer EXPOSES the Limits of AI SRE

Why watch

Alex Palcuie from Anthropic's AI Reliability team presents a technical deep-dive on using Claude for production incident response. The talk covers where LLMs excel (log parsing, anomaly detection, signal gathering) versus where they fail dangerously (confusing correlation with causation during complex incidents). Using John Boyd's OODA Loop framework and SAE autonomy levels, Palcuie analyzes real case studies including New Year's Eve fraud detection and Rust panic investigation, concluding with five practical patterns for integrating LLMs into incident management workflows. Targeted at SREs and platform engineers evaluating AI ops tools.