Miller's chapter on observability addresses a failure mode that is extremely common and rarely anticipated by first-time AI builders: model drift. A model trained on historical data performs well when it is deployed. But over time, the world changes — user behaviour shifts, new patterns emerge, the distribution of inputs changes — and the model's performance degrades. Without monitoring, this degradation is invisible until it has already caused significant damage. Observability is the discipline of building systems that make AI performance visible in real time: dashboards that track model accuracy over time, alerts that fire when performance drops below a threshold, logs that make it possible to audit why the model produced a specific output. Building observability into your AI system from the beginning is far easier than retrofitting it after a problem occurs.