The text discusses the inadequacy of traditional observability methods when applied to AI systems, highlighting a growing crisis in AI monitoring. Traditional monitoring tools fail to detect non-deterministic issues specific to AI, such as incorrect outputs and silent failures, which can lead to significant business disruptions and financial losses. Examples like DPD's customer service AI mishap and Air Canada's legal issues illustrate the real-world consequences of inadequate AI monitoring. To address these challenges, the text advocates for a shift towards AI-specific observability that focuses on evaluating AI interactions for correctness, understanding context, intent, and linking costs to business outcomes. The piece underscores the importance of developing comprehensive, real-time visibility into AI operations to ensure trustworthy and effective AI systems.