Skip to content
AI Training for Teams

Glossary

AI terms in plain language

Explained for companies, not for engineers. So your team can agree on what things mean and AI discussions stop being chaos.

AI basics

11 terms

  • AI-native process

    An AI-native process is designed from the start on the assumption that AI agents do a substantial part of the work. It is not a sped-up old procedure, it is a different one.

  • API key

    An API key is a unique security code that works like a password. It lets your company applications communicate safely and exchange data with other systems such as ChatGPT.

  • Copilot (AI assistant)

    A digital AI assistant built directly into the tools you use every day, such as Excel, Word or your CRM. It helps you work faster and make fewer mistakes.

  • Generative AI

    A type of AI that creates entirely new content from a user's instructions: text, images, spreadsheets, or even software code.

  • Machine learning

    Machine learning is the foundational branch of artificial intelligence. Systems learn to recognise patterns in data and improve with practice, without being programmed for each specific case.

  • NLP (Natural Language Processing)

    NLP (Natural Language Processing) is the AI technology that lets computers read, understand, analyse and generate human language.

  • No-code / Low-code

    Building software and automations by connecting blocks visually instead of programming. It lets ordinary employees automate their own work themselves.

  • Prompt

    A prompt is the text instruction you write to an AI such as ChatGPT or Claude in order to get the result you want from it. The quality of the prompt directly determines the quality of the answer.

  • Proof of Concept (PoC)

    A small-scale test of an idea. It lets a company confirm on a minimal budget whether a given AI technology is worth investing in properly.

  • Token

    A token is the "piece of a word" an AI uses to read and generate text. It is also the basic unit AI providers bill by, which makes it the unit your costs are measured in.

  • Token consumption

    Token consumption is how much text a team pushes through AI models. It is the main cost line in working with AI and a good indicator of how intensively the team uses it.

AI in practice

18 terms

  • Agentic engineering

    Agentic engineering is an approach where developers do not write code by hand but design the environment, rules and checks in which AI agents reliably produce it.

  • AI adoption ladder

    The AI adoption ladder is a proven, pragmatic way of introducing AI into a company in stages. It minimises the risk and maximises the return at each step.

  • AI skill

    An AI skill is a reusable, named procedure packaged so the AI invokes it itself at the right moment. The company writes it once and the whole team uses it.

  • AI without hype

    AI without hype is a pragmatic approach that ignores media bubbles and science-fiction scenarios entirely. It focuses only on what works today, saves time and makes money.

  • Automation

    Wiring AI into company processes so that repetitive tasks run by themselves, without a person stepping in. It saves time and removes the errors that come from tiredness.

  • CLAUDE.md

    CLAUDE.md is a text file in your project holding the rules for AI: how code is written here, what is forbidden, how things get tested. The AI reads it before every task.

  • Context engineering

    Context engineering is the systematic work of managing what an AI is given before a task: data, documents, examples of good output, and the rules of your company.

  • Custom GPTs

    Custom GPTs are specialised versions of ChatGPT you can tailor to specific tasks using your own instructions and your own company data.

  • Few-shot prompting

    A prompting technique where you include several concrete samples and examples in the text, showing exactly what the expected result should look like.

  • Harness engineering

    Harness engineering is designing the guardrails around an AI agent: automated tests, code review, CI/CD, monitoring and security checks that catch a mistake before a person does.

  • Chain of thought

    Making an AI write out its reasoning step by step before giving the final answer. It cuts the error rate dramatically on complicated tasks.

  • Integration platforms (Make / Zapier)

    Tools such as Make.com or Zapier that let you connect hundreds of different business applications to each other and automate moving data between them.

  • Monday morning test

    The Monday morning test is a simple way to judge whether AI training worked. It asks what an employee will actually do differently on Monday morning.

  • Return on investment in AI (ROI)

    The measure (Return on Investment) of how many euros, or how much saved time, comes back for every euro you put into AI tools and training.

  • Spec-driven development

    Spec-driven development means an exact brief exists before any code: the problem, the requirements, acceptance criteria and tests. The AI then builds from that rather than from guesswork.

  • Vibe coding

    Vibe coding is a way of building software where you describe what you need and the AI writes the code. It suits prototypes and personal tools, not production systems.

  • Workflow

    The series of steps a company has to go through to complete a given task. AI can speed those steps up, join them together, or take them over entirely.

  • Zero-shot prompting

    A technique for instructing AI where you put the question or the task to the model directly, without showing it any example of a solution first.

Risks and limits

6 terms

  • Bias (AI bias)

    Skew or partiality in an AI system. It arises when the AI learns from incomplete or unrepresentative data and consequently makes unfair decisions.

  • Data protection in AI

    The set of rules and settings that keep company data you put into AI tools from being misused or used to train public models.

  • EU AI Act

    The European Union law regulating the use of artificial intelligence. It sorts AI systems by level of risk and sets companies clear rules of the game.

  • GDPR and AI

    The rules and procedures that keep a company's use of AI compliant with the European General Data Protection Regulation.

  • Hallucination (AI)

    When an AI model confidently generates information that is wrong or entirely invented. It looks like a fact, but it is not a fact.

  • Human in the loop

    A way of working where the AI does the heavy lifting (the draft, the analysis) but the final check and approval always stay in the hands of a live person.

Technology and models

15 terms

  • AI agent

    An AI agent is a system that does not only generate text but can take concrete steps on its own: searching the web, sending emails, or changing records in your CRM.

  • API (Connecting systems)

    A bridge between different pieces of software that lets them exchange data and work together without a person in the middle. The key to any automation in a company.

  • Claude Code

    Claude Code is Anthropic's tool that lets AI reach your files and projects directly. Instead of copying code back and forth, the agent works in the folder and proposes changes for approval.

  • Cloud AI

    AI that does not run on your computer but on the provider's powerful servers (Microsoft, Google, OpenAI and the like). You reach it over the internet.

  • Context window (AI memory)

    The short-term working memory of an AI model. It sets the maximum amount of text and data the AI can process within a single request.

  • Embeddings (Vectors)

    The way an AI turns words and sentences into numbers (vectors). That is what lets it grasp the meaning of text and find connections between different documents.

  • Fine-tuning

    Fine-tuning is additional training of an existing AI model on your specific company data so it learns your particular tone, style or jargon.

  • ChatGPT Enterprise

    The version of ChatGPT built for companies. Its main advantage is a guarantee that your company data will not be used to train further AI models, plus team management tools.

  • LLM (Large Language Model)

    An LLM (Large Language Model) is a type of AI trained on an enormous amount of text. That training lets it read, understand and write text that is hard to tell apart from a human's.

  • Managed AI agent

    A managed AI agent runs on the provider's infrastructure instead of your laptop. It handles long and repeated tasks, keeps its context, and needs no computer left switched on.

  • On-premise AI (Local AI)

    Local AI that runs directly on company computers or servers, so sensitive company data never leaves your building at all.

  • Open-source AI

    AI with open source code. You can download it, run it on your own servers, and pay no licence fees to giants like OpenAI or Google.

  • RAG (Retrieval-Augmented Generation)

    RAG is an approach that connects an AI model to your own company documents. It makes the AI answer strictly from your internal data instead of inventing things.

  • System prompt

    The underlying, invisible instruction that defines an AI's role, tone and boundaries before you ever ask it your first question.

  • Vector database

    A special kind of database that stores text as numbers. It lets AI find connections across thousands of company PDFs and contracts almost instantly.

The fastest next step

Want your team speaking one language?

Book a 15-minute call. We look together at whether the workshop makes sense for your company and how we could align your processes.

Book a consultation 15 minutes, free, no commitment.