Application Performance Monitoring refers to tooling that observes the runtime behavior of production applications: request latency distributions, error rates, throughput, database query performance, external service call performance, and end-to-end transaction traces. Major APM vendors include Datadog, New Relic, Dynatrace, AppDynamics, and open-source alternatives including OpenTelemetry- based stacks.
APM's role has expanded substantially with the complexity of modern distributed systems. A single user request might traverse a dozen microservices; identifying which service contributes to slow response times requires distributed tracing that stitches together spans across service boundaries. This is the primary technical evolution from traditional "application monitoring" to modern APM.
For AI-deployed systems, APM is becoming essential for reasoning about agent behavior. The equivalent of a "slow query" in an agent system is a slow or failing tool call, and tracing those calls across the agent's execution is now a standard APM feature. The overlap between APM and agent-specific observability tools (LangSmith, Langfuse, Helicone) is likely to consolidate over the next few years.