Which properties of modern systems broke classic monitoring?

Cloud-native architectures, microservices, containers and Kubernetes, distributed applications and a high rate of change through CI/CD — you can no longer directly observe what is happening inside the system.

Each property removes one assumption the old model relied on:

• Cloud-native and distributed applications — there is no single machine to log into; the "system" is a conversation between many.
• Microservices — a failure emerges from the interaction of components, each of which is individually healthy.
• Containers and Kubernetes — the unit you would have investigated is ephemeral. By the time you look, the pod that failed has been replaced and its local state is gone.
• High change frequency (CI/CD) — the thing that broke may have existed for twenty minutes, so no one has experience with its failure modes.

The consequence: errors arise across many components, classic monitoring shows only symptoms — a queue is growing, latency is up — and root-cause analysis gets steadily harder as the number of possible culprits rises.

Tip: the core statement worth keeping is "you can no longer directly observe what happens inside the system." Everything about observability is a response to that one sentence.