Observability combines three kinds of data. Logs record discrete events, metrics track numbers over time such as error rates and response times, and traces follow a single request through multiple services. Together they answer what is broken, where, and why.
Good observability includes alerts tied to real user impact, dashboards for the system's key health indicators, and, for AI systems, tracking of model usage, cost, and quality. Alerts that fire constantly get ignored, so tuning them is part of the work.