The Executive's AI Vocabulary Guide: 40 Terms You Need to Know

1. Large Language Model (LLM)

The engine behind ChatGPT, Claude, etc. A program trained on massive amounts of text that can generate human-like responses. Think of it as a very sophisticated autocomplete that understands context.

2. Generative AI

AI that creates new content — text, images, code, audio. This is the category that includes ChatGPT, Midjourney, and most tools in this guide. Distinguished from AI that classifies or predicts.

3. Prompt

The instruction you give to an AI. Better prompts → better output. This is a skill worth developing. See our AI Prompt Cheat Sheet for templates.

4. Context Window

How much text an AI can "see" at once. Measured in tokens (~words). ChatGPT: 128K tokens. Claude: 200K tokens. Bigger = can handle longer documents.

5. Hallucination

When AI generates confident-sounding information that is factually incorrect. It doesn't "know" it's wrong. This is why verification matters, especially for business decisions.

6. Fine-tuning

Training a general AI model on your specific data to make it better at your tasks. Like hiring a generalist consultant and teaching them your industry.

7. RAG (Retrieval-Augmented Generation)

A technique where AI searches your own documents before answering, reducing hallucinations. This is how Notion AI answers questions about your workspace.

8. Token

The unit AI uses to measure text. Roughly 1 token = 0.75 words. Important for understanding pricing and context window limits.

9. API (Application Programming Interface)

A way for software to talk to AI programmatically. When a company "integrates AI," they're usually using an API from OpenAI, Anthropic, or Google.

10. Foundation Model

The base AI model (GPT-4, Claude, Gemini) before it's customized. Think of it as the raw material that gets shaped into specific products.

11. AI Copilot

AI that assists humans rather than replacing them. Microsoft's "Copilot" branding popularized this — the AI helps you do your job faster, you stay in control.

12. AI Agent

AI that can take actions autonomously — browse the web, send emails, manage workflows. More independent than a copilot. Still emerging technology in 2026.

13. Prompt Engineering

The skill of writing effective AI prompts. Not as technical as it sounds — mostly about being clear, specific, and providing good context.

14. AI Governance

Policies and rules for how your organization uses AI. Covers data privacy, acceptable use, quality control, and compliance. Every company needs this.

15. Shadow AI

Employees using AI tools without IT approval. Like shadow IT, but for AI. A growing concern for compliance and data security teams.

16. AI ROI

Return on investment from AI tools. Measured in time saved, output quality improvement, or cost reduction. Hard to measure precisely, but time-tracking helps.

17. Responsible AI

Designing and using AI ethically — considering bias, fairness, transparency, and societal impact. Not just a PR concept; increasingly a regulatory requirement.

18. AI Literacy

Understanding what AI can and can't do, how to use it effectively, and when to trust its output. The most important professional skill of 2026.

19. Multimodal AI

AI that handles multiple types of input — text, images, audio, video. GPT-4o and Gemini are multimodal. Important for workflows that involve different media types.

20. Vertical AI

AI built for a specific industry (legal AI, healthcare AI, financial AI). Often more accurate for specialized tasks than general-purpose models.

21. Transformer

The AI architecture behind modern language models. You don't need to understand how it works — just know that it's what makes models like GPT-4 and Claude possible.

22. Training Data

The text/images/code an AI was trained on. Determines what the AI "knows." Training data has a cutoff date, which is why AI can't always answer about recent events.

23. Inference

When AI generates a response. "Running inference" = using the AI. This is what you pay for with API usage.

24. Embedding

A way of converting text into numbers so AI can compare meanings. This is how AI search works — finding things that are semantically similar, not just keyword matches.

25. Latency

How long it takes AI to respond. Smaller models are faster (lower latency). Matters when AI is part of a real-time workflow.

26. Open Source vs Closed Source

Open source models (LLaMA, Mistral) share their code publicly. Closed source (GPT-4, Claude) don't. Open source = more customizable. Closed source = generally more capable.

27. Guardrails

Safety measures built into AI to prevent harmful output. Why Claude won't help you write a phishing email. Important for enterprise deployment.

28. Temperature

A setting that controls AI creativity. Low temperature = more predictable, factual. High temperature = more creative, varied. Most professional tasks want low-medium temperature.

29. Zero-shot vs Few-shot

Zero-shot: asking AI to do something with no examples. Few-shot: giving it 2-3 examples first. Few-shot prompting is a simple technique that dramatically improves output quality.

30. Benchmark

A standardized test for comparing AI models. Examples: MMLU, HumanEval. Useful for comparing models, but real-world performance often differs from benchmark scores.

31. AI-Native

A company or product built with AI at its core, not bolted on after. Perplexity is AI-native search. Google adding AI to search is not AI-native.

32. Compound AI System

Multiple AI models working together. Instead of one big model doing everything, specialized models handle different parts of a workflow.

33. Synthetic Data

Fake data generated by AI to train other AI. Used when real data is scarce, sensitive, or expensive to collect.

34. AI Alignment

Making sure AI does what humans actually want, not just what they literally asked for. A major research focus for safety.

35. Edge AI

Running AI on local devices instead of cloud servers. Faster, more private. Apple Intelligence on iPhone is an example.

36. Agentic AI

AI systems that can plan, reason, and execute multi-step tasks autonomously. The next frontier beyond chatbots.

37. Model Distillation

Creating a smaller, faster AI from a larger one. The smaller model learns to mimic the larger model's behavior. Why some AI is getting cheaper.

38. Mixture of Experts (MoE)

An architecture where different parts of the AI activate for different tasks. More efficient than having the whole model process every request.

39. Constitutional AI

Training AI with explicit principles about what's helpful, harmless, and honest. Anthropic's approach to AI safety.

40. AI Regulation

Government rules for AI. The EU AI Act is the most comprehensive. Understanding this matters for compliance, risk management, and strategic planning.