
Explore Discover, the Knowledge Hub, playbooks, and the Operator Vault.
Glossary
Generative AI is a class of artificial-intelligence systems designed primarily to produce new content — such as text, images, audio, code, or other structured outputs — conditioned on prompts, context, or other inputs. Contemporary generative systems are often built on foundation models (including large language models), but generative AI is not identical to AI as a whole, to predictive/discriminative AI, or to agentic systems that plan and act with tools.
Enterprise programs frequently label every model initiative “generative AI,” which collapses distinct design and risk choices. Generative systems change the work pattern from classifying or forecasting to drafting and synthesizing artifacts at scale — which raises questions of quality, IP, hallucination, evaluation, and workflow redesign that differ from classical predictive models.
Useful distinctions: predictive AI estimates labels or scores; foundation models are broadly adaptable pretrained bases; LLMs are a major text-centric foundation-model family; generative AI names the content-producing application class; agentic systems add goal-directed action loops over tools and environments. A chat assistant may be generative without being a full agent.
Hypothetical illustration: A product organization deploys a generative drafting assistant for customer emails using a foundation-model API with retrieval. Separately, a fraud team continues to run a predictive risk-scoring model. Treating both as the same “AI program” obscures different evaluation metrics, failure modes, and governance controls — generative quality and hallucination risk versus precision/recall on fraud labels.
A short path from this concept into related research and practical resources.