AI glossary for boards and leaders
AI terminology moves quickly. This glossary explains the terms directors and leadership teams are increasingly likely to encounter. AI terminology moves quickly. This glossary explains the terms directors and leadership teams are increasingly likely to encounter in board papers, management discussions, vendor proposals and conversations about AI strategy, adoption and risk.
Browse by topic
All terms
Each term includes a definition, what matters for boards and leaders, and the questions they need to think about.
A
- A2A (Agent-to-Agent protocol) A protocol designed to allow AI agents built on different systems or platforms to communicate, exchange information and coordinate work. Read more
- Agent identity and authorisation The mechanisms used to establish which AI agent is acting and what that agent is permitted to access or do. Read more
- Agent Skills Reusable packages of instructions, processes and supporting resources that teach an AI agent how to perform a particular task or type of work. Read more
- Agentic AI These AI systems can pursue goals, make intermediate decisions, use tools and perform multiple steps with varying levels of autonomy. Read more
- Agentic workflow A workflow in which AI decides at least some of the steps required to achieve an outcome rather than following an entirely predetermined sequence. Read more
- AI agent An AI-enabled system that can pursue a goal, use tools or systems and take actions on behalf of a person or organisation. Read more
- AI FinOps The emerging discipline of measuring, forecasting and managing the cost of AI services, models and infrastructure. Read more
- AI gateway A control layer through which an organisation can route AI requests to different models while applying common security, logging, cost and governance controls. Read more
- AI observability The ability to see and understand how an AI system is behaving in operation, including its outputs, actions, errors, costs and use of tools. Read more
- Automation bias The tendency for people to accept or over-rely on automated recommendations even when there is reason to question them. Read more
C
- Computer use An AI capability that allows an agent to interact directly with software interfaces by viewing screens, clicking, typing and navigating applications. Read more
- Context engineering The design and management of the information, instructions, memory, tools and resources made available to an AI system. Read more
- Context rot A term used to describe deteriorating AI performance as its working context becomes crowded with outdated, duplicated, conflicting or irrelevant information. Read more
- Context window The amount of information an AI model can work with at one time, including instructions, conversation history, documents and other supplied context. Read more
- Copilot / AI copilot A term commonly used for AI that assists a person while the person remains responsible for directing the work and taking the final action. Read more
- Cowork A term introduced by Anthropic for an AI experience designed around delegating substantial, multi-step work rather than simply exchanging chat messages. Read more
D
E
F
G
- Generative AI AI capable of generating new content such as text, images, audio, video or code in response to instructions or context. Read more
- Grounding Providing an AI system with relevant, trusted information to anchor its responses or actions. Read more
- Guardrails Technical or procedural controls designed to constrain AI behaviour or prevent undesirable actions or outputs. Read more
H
- Hallucination An AI-generated statement or output that appears plausible but is false, unsupported or invented. Read more
- Human-at-the-helm An emerging expression describing human responsibility for direction, boundaries and accountability when AI performs work with greater autonomy. Read more
- Human-in-the-loop (HITL) A design in which a person participates directly at one or more points in an AI-enabled process, often by reviewing or approving an output or action. Read more
- Human-on-the-loop An oversight model in which AI performs work without approval at every step while a person supervises performance and can intervene. Read more
I
K
L
M
- MCP (Model Context Protocol) An open protocol that provides a standard way for AI applications and agents to connect with external tools, systems and sources of information. Read more
- Model drift A deterioration or change in AI performance as the environment, data, users or conditions around the system change over time. Read more
- Model routing Automatically selecting which AI model should handle a particular task. Read more
- Multi-agent system An AI system in which multiple agents perform different roles or coordinate to complete a task. Read more
- Multimodal AI AI capable of processing or generating more than one type of information, such as text, images, audio and video. Read more
O
P
- Probabilistic AI AI whose outputs or behaviour are based on probabilities rather than entirely predetermined rules, meaning results can vary across apparently similar tasks. Read more
- Prompt engineering The practice of designing and refining instructions given to an AI model to improve the quality or consistency of its output. Read more
- Prompt injection An attack or failure mode where instructions encountered by an AI system cause it to ignore, reinterpret or conflict with its intended instructions. Read more
- Proprietary / closed model An AI model whose weights and internal model assets are controlled by the provider and typically accessed through the provider's service or an authorised platform. Read more
R
- RAG (Retrieval-Augmented Generation) A technique where an AI system retrieves relevant information from an external source and supplies it to a model when generating a response. Read more
- Reasoning model An AI model designed to spend additional computational effort working through complex problems before producing a response. Read more
- Reasoning tokens Tokens used internally by some reasoning models while working through a problem before producing the visible answer. Read more
- Red teaming Deliberately trying to make an AI system fail, behave unsafely or operate outside its intended boundaries. Read more
S
- Sandboxing Running an AI system or agent inside a restricted environment that limits what it can access or affect. Read more
- Shadow AI AI being used inside an organisation without appropriate visibility, approval or governance. Read more
- Synthetic data Artificially generated data created to resemble characteristics or patterns found in real data. Read more
T
- Token A unit into which AI models break information when processing input and generating output. Read more
- Tokenomics In AI, tokenomics describes the economics of token consumption and how AI usage translates into operating cost. Read more
- Tracing Recording the sequence of model calls, tool calls, intermediate steps and actions taken by an AI system. Read more
U
Can't find the term you need?
Tell us what your board or leadership team keeps hearing, and we'll consider adding it.
