# What is Monitoring?

Also called system monitoring.

Monitoring is the continuous collection and evaluation of signals about a running system to determine whether it is behaving as expected. It focuses on known indicators such as error rate, latency, saturation, and traffic, checked against defined thresholds. Monitoring answers whether something is wrong, while broader observability practice is concerned with explaining why.

Monitoring is built around signals chosen in advance. Standard framings include the four golden signals of latency, traffic, errors, and saturation, and the utilization, saturation, and errors method for resources. Both exist to keep dashboards focused, because a system emitting thousands of undifferentiated metrics is harder to reason about during an incident than one with a small set of well chosen indicators.

The most important distinction is between symptom based and cause based monitoring. Symptom monitoring tracks what users experience, such as failed requests or slow responses. Cause monitoring tracks internal conditions such as processor usage or queue depth. Symptoms are what should wake someone up, because they correlate with real harm, while causes are what accelerate diagnosis once attention is already on the problem.

Coverage gaps are usually about absence rather than error. Systems reliably notice things that fail loudly and reliably miss things that stop happening: a scheduled job that never starts, a queue that stops receiving work, a webhook sender that goes quiet. Monitoring for expected activity within a time window catches this class of silent failure, which error monitoring alone never will.

Agent systems add indicators that traditional monitoring does not cover. Beyond uptime and latency, operators track tool call failure rates, how often runs end without completing their task, escalation and fallback frequency, and metered usage per run. A platform can be entirely healthy by infrastructure metrics while the work it produces has quietly degraded, so quality signals belong alongside system signals.

## Key points

- Continuous checking of predefined signals against thresholds.
- Latency, traffic, errors, and saturation are the common core.
- Symptom monitoring pages; cause monitoring diagnoses.
- Watch for absence of expected activity, not only for errors.
- Agent systems need quality signals beside infrastructure ones.

## In practice

An operations dashboard shows request error rate, ninety fifth percentile latency, queue depth, and agent task completion rate side by side. One morning error rate and latency look normal while completion rate drops from ninety four percent to sixty percent. Nothing is technically failing, but a tool integration has started returning empty results, so the runs finish without doing useful work. Only the quality signal reveals it.

## Related terms

- [Observability](/en/glossary/observability)
- [Alerting](/en/glossary/alerting)
- [Service Level Objective](/en/glossary/service-level-objective)
- [Uptime](/en/glossary/uptime)
- [Incident](/en/glossary/incident)

[Back to the AI Glossary](/en/glossary)