AI at work
AI at work: a plain-English glossary
The AI terms people actually run into at work, explained in plain language with the workplace angle that matters — no hype, no jargon. Learn the word, then practice the skill in a guided lab with the AI tool you already use.
- Prompt
- The instructions you give an AI tool. A good workplace prompt names your role, the task, the useful context, the output you want, and what to leave out.
- Context
- The background facts you give the AI so its answer fits your situation — your role, the audience, the source material, and any rules to follow.
- Context window
- How much text an AI can consider at once (your prompt plus its reply). Long documents can exceed it, so summarize or paste only the relevant parts.
- Large language model (LLM)
- The kind of AI behind tools like ChatGPT, Claude, and Copilot. It predicts likely text, which is why it can sound confident even when it's wrong.
- Hallucination
- When an AI states something false as if it were fact — an invented figure, source, policy, or name. Always verify important claims before you use them at work.
- Token
- The small chunks of text an AI reads and writes (roughly a word or part of a word). Limits and costs are usually measured in tokens.
- System prompt
- Hidden instructions that set an AI assistant's behavior and boundaries before your message. Your own prompt still shapes each answer.
- Grounding
- Giving the AI real source material (a document, notes, data) so its answer is based on facts you provided rather than its general training.
- Retrieval-augmented generation (RAG)
- A setup where the AI first looks up relevant documents, then writes an answer grounded in them. It reduces hallucination when it works well, but the sources still need checking.
- Zero-shot vs few-shot
- Zero-shot means you ask with no examples; few-shot means you include one or two examples of the output you want. Examples usually improve results.
- Chain-of-thought
- Asking the AI to reason step by step before answering. Useful for multi-step tasks, but the steps can still be wrong, so review the conclusion.
- Temperature
- A setting some tools expose that controls randomness. Lower is more focused and repeatable; higher is more varied and creative.
- Guardrails
- The limits that keep AI use safe at work — what not to paste, when a human must review, and which tasks the AI should not decide on its own.
- Prompt injection
- A trick where hidden text in a document or web page tries to make the AI ignore your instructions. Treat content the AI reads as data, not commands.
- PII (personally identifiable information)
- Details that identify a person — names, emails, phone numbers, account or ID numbers. Remove it before pasting into an AI tool unless the tool and use are approved.
- Shadow AI
- Employees using AI tools their company hasn't vetted, often pasting in work data. The fix is training people to use approved tools safely — not banning AI.
- Human in the loop
- Keeping a person responsible for reviewing and approving AI output before it's used, especially for anything sensitive, high-impact, or authority-bound.
- Reusable prompt (prompt template)
- A prompt you save and run again for a recurring task, with blanks for the details that change. Reuse is where AI turns from a novelty into real time saved.
- Model fit
- Choosing the right kind of AI help for the work: a general chat model for drafts, a reasoning model for tradeoffs, a search-connected tool for current facts, or an approved enterprise tool for sensitive data.
- Output review
- Checking AI output before you use it — for accuracy, missing information, tone, privacy, and whether a human should sign off. The habit that makes AI safe at work.
Learn the word, then the skill
Knowing the terms is step one
AI Lunchroom turns these ideas into practice on your real work — with the AI tool you already use. Free to start, no credit card.