Review Latency

Latency is the elapsed time between starting a request or event and receiving the relevant response or result.

In an AI system, latency may include network travel, retrieval, model inference, tool calls, safety checks, and application processing. The total delay is what the user and workflow experience.

A technically capable system can still fail in practice if responses arrive too slowly for the task, interrupt conversation flow, or create queues elsewhere in the process.

A call-center assistant takes twelve seconds to suggest an answer, causing agents to stop waiting and return to manual search.

Latency is not only the model's generation speed. Every handoff in the complete system can contribute to the delay.

Latency is how long the system makes the work wait.

* [[inference|Inference]] * [[friction|Friction]] * [[workflow|Workflow]] * [[bottleneck|Bottleneck]] * [[service-level-agreement|Service-Level Agreement]]

* [NIST Computer Security Resource Center, Glossary: Latency](https://csrc.nist.gov/glossary/term/Latency) — NIST-sourced definition of latency as a processing time delay. * [Google, *Site Reliability Engineering: Service Level Objectives*](https://sre.google/sre-book/service-level-objectives/) — operational treatment of latency as a user-facing service indicator measured across request distributions.

* Map delay through [[workflow|Workflow]] and distinguish it from a persistent [[bottleneck|Bottleneck]].