AI Glossary
A shared language for AI.
Plain-language definitions for better conversations and decisions.
What do these AI terms mean?
Generative AI
- AI that produces content such as text, images or code in response to input. Generated output may be inaccurate.
Prompt
- Instructions and context given to an AI system to guide its response.
Prompt engineering
- The practice of structuring and refining instructions, context and constraints for a specific task.
Large language model (LLM)
- A model trained on large amounts of data to process and generate language. Fluent output is not proof of correctness.
AI agent
- A system that uses an AI model with tools to take steps toward a task. Its permissions and actions should be explicitly controlled.
AI automation
- A workflow that uses AI to carry out one or more process steps, with appropriate checks and exception handling.
Hallucination
- An AI-generated statement that appears plausible but is unsupported, false or inconsistent with the source.
Retrieval-augmented generation (RAG)
- An approach that retrieves relevant information and provides it as context for generation. Retrieval does not guarantee a correct answer.
Human in the loop
- A workflow with explicit human review or approval at defined points.
AI governance
- Rules, responsibilities and review processes for how an organization selects and uses AI.
AI literacy
- Practical understanding of AI capabilities, limitations and responsible use.
Evaluation
- A structured check of an AI system’s outputs or behavior against relevant criteria and representative examples.
Put the vocabulary in context
How do these terms connect in a business workflow?
A prompt gives instructions and context to an AI model. A workflow surrounds that interaction with inputs, review and a next action. Retrieval can supply approved reference material; evaluation checks whether the resulting answer meets the task’s requirements. Governance defines who owns those choices and which uses are acceptable.
An agent adds the ability to take steps using tools. That makes permissions and approval boundaries particularly important. Calling a workflow an agent does not establish that it is more useful, more autonomous or safer than a simpler design.
Related resources
