Data center anomaly detection with AI agents

This post details a hybrid approach to using AI agents for data center anomaly detection. It describes using agents offline to build deterministic anomaly detection pipelines (feature engineering, model training, threshold setting) and then using agents online to reason about flagged anomalies, explain causes, and recommend actions. The post highlights the trade-offs between agents as builders of tools (deterministic output) and agents as tools themselves (reasoning over live data), emphasizing the need for human review in critical infrastructure decisions.