The terms you'll hear — and what they actually mean.
Prompt Tier 1
The instruction you give an AI. Quality of output = quality of prompt.
✓ Most transferable AI skill you can build.
Hallucination Tier 1
When AI confidently states something completely made up — delivered as fact.
⚠️ Always verify before AI output leaves your org.
LLM Tier 1
Large Language Model — the technology behind most AI writing tools.
✓ Explains why AI is good at some things and not others.
Agent / Agentic AI Tier 1
AI that takes actions — not just answers questions. Fundamentally different risk level.
⚠️ Standard guardrails don't apply.
Copilot Tier 1
Means at least six different things. Always ask which Copilot is being discussed.
✓ M365 users may already have Copilot access.
Claude / GPT / Gemini Tier 1
AI models from Anthropic, OpenAI, and Google.
✓ You may already be using one without knowing it.
RAG Tier 2
How "talk to your documents" tools work. Your data quality determines results.
✓ Clean data = dramatically better AI results.
Token Tier 2
The unit AI models use to measure text — roughly ¾ of a word.
✓ Model out costs before signing volume agreements.
Vendor Lock-in Tier 2
When switching vendors becomes so painful you feel trapped.
⚠️ Ask "how do I export my data?" before signing.
AI Washing Tier 2
Vendors adding "AI-powered" without meaningfully changing the product.
⚠️ Ask vendors to demonstrate what the AI actually does.
Data Residency Tier 2
Where your data is physically stored. Regulated industries have specific requirements.
⚠️ Required vendor question for regulated industries.
Acceptable Use Policy Tier 2
Written policy defining how staff may use AI — what's allowed, what data is off-limits.
✓ Write it before you need it.