Home / AI glossary
Enterprise AI glossary
Plain-language definitions for teams planning AI automation. These definitions describe behavior without implying a guaranteed result.
- AI agent
- Software that uses a model plus tools, instructions, and state to complete a bounded task. It still needs permissions, failure handling, monitoring, and escalation.
- Voice AI agent
- A system combining telephony, speech recognition, a conversational model, business tools, and speech synthesis to handle a phone interaction.
- Large language model (LLM)
- A model trained to predict and generate language. It can support complex workflows but does not guarantee factual correctness.
- Retrieval-augmented generation (RAG)
- A pattern that retrieves material from an approved knowledge source and supplies it to a model when producing an answer.
- Hallucination
- A plausible-sounding output that is unsupported, incorrect, or invented. Retrieval can reduce some errors but does not eliminate them.
- Guardrail
- A technical or procedural control constraining inputs, outputs, tool use, permissions, or escalation behavior.
- Evaluation
- A repeatable test of system behavior using defined examples, scoring rules, thresholds, and failure categories.
- Human in the loop
- A workflow where a person reviews, approves, corrects, or takes over at a defined point.
- Latency
- The time between a request and usable response, including delays introduced by models, telephony, and integrations.
- Pilot
- A contained implementation that tests feasibility, reliability, adoption, cost, and measurable outcomes before wider rollout.
- Baseline
- The pre-implementation value of a metric measured using the same definition and population used after launch.
- Model drift
- A change or decline in performance as inputs, user behavior, business rules, or the environment change.