Brian Brazil discusses the evolution from machine-centric monitoring to service-oriented monitoring in modern computing environments. Traditionally, each machine had a single purpose with monitoring focused on machine metrics like CPU usage, which often resulted in irrelevant alerts and pager fatigue due to false positives and unactionable alerts. As systems grow, this approach becomes less effective, as it does not align with users' concerns, which are more focused on service-level metrics such as availability, latency, and error rates. Brazil suggests a shift to aggregating machine-level metrics to the service level for more meaningful alerts, emphasizing the need to monitor services in the context of their performance and reliability rather than individual machine metrics. This change is particularly relevant in microservice architectures where machines host multiple services, making traditional monitoring methods inadequate and noisy.