The GenAI for Business Glossary

47 key terms of generative AI for business, defined in plain managerial language: from prompt engineering and RAG to loop engineering, evals, MCP, and AI sovereignty.

• Model Parameters
  Parameters are the adjustable "knobs" inside an AI model, the numerical values it learns during training. A large frontier model (the GPT, C…
• Fine-Tuning
  Fine-tuning means taking a general AI model and retraining it on smaller, specialized datasets. It teaches the model new vocabulary, tone, o…
• Temperature
  Temperature controls how creative or predictable a model's responses are: Low temperature (0.1 to 0.3): Deterministic and factual. Ideal for…
• Top-K Sampling
  Top-K limits how many possible next words (tokens) the model considers. For example, with K = 50, the model only looks at the 50 most likely…
• Retrieval-Augmented Generation (RAG)
  RAG connects an AI model to external information sources (like documents or databases). The model first retrieves relevant facts (like a "li…
• Context Engineering
  Context engineering is about designing the environment the AI uses to think and respond. It includes what background information, examples, …
• The Five A's of Applied GenAI
  A framework to understand how GenAI transforms work at different levels: Stage Focus Example Access Using foundation models (the GPT, Claude…
• The EDGE Framework
  A strategic model for capturing GenAI's value: Dimension Meaning Goal E: Efficiency Streamline and automate processes Cost and time savings …
• Prompt Engineering
  The craft of designing inputs (prompts) to guide AI models toward better results. Good prompts are clear, specific, and structured. They may…
• Tokenization
  Before a model processes text, it breaks it into tokens, small units (words, subwords, or symbols). For example, "marketing" might become [m…
• Scaling Laws
  Scaling laws describe how AI performance improves as you increase model size, data, and compute power, usually in predictable ways. The rule…
• Agents
  AI agents can understand goals, plan tasks, and execute them autonomously using tools or APIs. They go beyond chatbots, for example, an agen…
• API (Application Programming Interface)
  An API is a bridge that lets software systems talk to each other. For Generative AI, APIs allow apps or platforms to send a prompt to a mode…
• Playground
  A playground is a sandbox environment where users can experiment with AI models safely. Platforms like OpenAI Playground, Google AI Studio, …
• Zero-Shot, One-Shot, and Few-Shot Learning
  These terms describe how much prior example input a model receives for a task: Zero-Shot: The model is asked to perform a task with no examp…
• Chain of Thought (CoT)
  CoT prompting asks the model to reason step-by-step rather than jump to a conclusion. For example: "Let's think step by step." This encourag…
• Tree of Thought (ToT)
  An evolution of Chain of Thought. Here, the AI explores multiple reasoning paths, compares them, and picks the best outcome, like branching …
• Self-Consistency
  Instead of relying on one output, the model generates several answers to the same prompt, then chooses or averages the most consistent one. …
• ReAct (Reason + Act)
  ReAct combines reasoning with actions. The model thinks, then acts, such as searching the web, running code, or calling another tool and the…
• Governance Framework (for GenAI)
  AI governance refers to the rules, structures, and oversight that ensure safe, fair, and legal AI use. A good framework covers: Data privacy…
• Synthetic Data
  Synthetic data is artificially generated rather than collected from real users. It mimics real-world patterns while protecting privacy and a…
• Hallucination
  A hallucination occurs when an AI model produces confident but false or fabricated information. It can happen due to incomplete training dat…
• Scaling Up vs. Scaling Out
  Scaling up means using bigger models or hardware (more parameters, GPUs, or memory). Scaling out means distributing workloads across multipl…
• Token Limit / Context Window
  Every model has a maximum "context window", the total number of tokens (words and symbols) it can process at once. For example, current GPT …
• Multimodality
  A multimodal model can process more than one type of data, text, image, audio, video, or code and combine them in output. For example, descr…
• Workflow Automation
  The use of AI to connect apps and automate repetitive tasks. Platforms like Zapier, n8n, and Make.com allow users to trigger AI actions (gen…
• Prompt Chaining
  Breaking a complex task into a series of smaller prompts where each step feeds the next. Example: Summarize a long report. Extract key insig…
• Embeddings
  Embeddings are numerical representations of text that capture meaning and similarity. They allow the AI to understand relationships between …
• Latent Space
  Latent space is the model's internal map of concepts, a multidimensional representation of ideas learned during training. In this space, sim…
• Diffusion Model
  A type of generative model that starts with random noise and gradually refines it into a meaningful image or sound. Used in image and video …
• Explainability (XAI)
  Explainable AI helps users understand why a model made a decision or produced an answer. It matters for trust and compliance, especially in …
• Model Alignment
  Model alignment ensures AI behaves according to human values, safety, and intent. It's achieved through training methods like reinforcement …
• Data Moat
  A data moat is a competitive advantage built from proprietary data that others can't easily copy. In GenAI, the more unique and high-quality…
• Human-in-the-Loop (HITL)
  A design where humans remain part of the AI process, reviewing, correcting, or approving outputs. HITL ensures quality, reduces risk, and he…
• GenAI Operating Model
  A GenAI operating model defines how an organization manages AI across functions, covering structure, workflows, ownership, and governance. I…
• Harness Engineering
  The discipline of designing the runtime system wrapped around a model: the agent loop, tool definitions, memory and state, permissions and s…
• Loop Engineering
  The discipline of designing systems that prompt the AI instead of prompting it yourself: autonomous cycles of act, observe, verify, and retr…
• Graph Engineering
  Structuring agentic work and organizational knowledge as explicit graphs: workflow graphs (pipelines, routing, orchestrator-workers, evaluat…
• GraphRAG
  A retrieval technique that builds a knowledge graph from documents (entities, relationships, community summaries) and retrieves along its ed…
• Model Context Protocol (MCP)
  An open standard, originated by Anthropic and adopted across the industry, through which AI agents connect to tools and data sources via a c…
• Evals (Evaluations)
  Systematic, repeatable tests of AI system quality: a set of real task examples with known good outputs, scored automatically or by an LLM ju…
• Reasoning Model / Test-Time Compute
  A model trained to "think" before answering, spending additional computation at inference time on internal reasoning steps. Scaling this tes…
• Open-Weight Model
  A model whose trained weights are published for download (e.g., Kimi, DeepSeek, Llama, Qwen), allowing self-hosting, fine-tuning, and inspec…
• Forward Deployed Engineer (FDE)
  A role, invented at Palantir and adopted across the AI industry, that embeds an engineer inside a customer's operation to scope real workflo…
• AI Sovereignty
  The capacity of an organization or country to use AI on its own terms: access that cannot be revoked by a foreign government or single vendo…
• Coding Agent
  An AI system that pursues a software goal with its own hands: it reads and edits a codebase, runs commands and tests, observes the results, …
• Permission Modes
  The dial on an agentic tool that sets how much it may do without asking: typically plan only (propose, touch nothing), ask for approval (the…