AI glossary — plain and simple
What is a prompt, RAG, a hallucination or the AI Act? Jargon-free explanations focused on what the terms mean for an ordinary business.
AI (artificial intelligence)
Software that handles tasks requiring human-like reasoning — understanding text, making decisions, creating content. For small businesses this most often means language models (ChatGPT, Claude) that read and write text: answering customers, sorting e-mails, drafting documents. AI services for businesses →
AI Act
The European AI regulation (2024/1689) — the world's first comprehensive AI law, directly applicable across the EU. Obligations phase in gradually: bans and AI literacy since February 2025, labelling of chatbots and AI content from 2 August 2026, high-risk system rules from December 2027. What companies must comply with from August 2026 →
AI agent
AI that doesn't just answer but performs tasks on its own — resolves an inquiry, creates a CRM record, sends an e-mail, escalates a problem to a human. Unlike a plain chatbot it works across multiple systems at once and makes decisions based on the situation. Custom AI agents →
AI audit
An independent assessment of where AI makes sense in your company — and where it doesn't. Includes process mapping, ROI calculations, review of existing AI projects and systems, and an AI Act compliance check. The output is a concrete plan with numbers, not a deck full of buzzwords. AI audits from €599 →
AI content watermarking
A machine-readable marker embedded in AI-generated content (images, videos, audio) that identifies it as AI-made. The AI Act requires it, with a transition period until 2 December 2026 — major providers already add watermarks to their outputs. AI content obligations →
AI literacy
An obligation under Article 4 of the AI Act: a company using AI must ensure employees can handle it appropriately for their role — knowing what may be entered into AI tools, how to verify outputs and where the limits are. In force since February 2025 for every company, regardless of size. How to comply in a single afternoon →
API
An interface through which programs talk to each other. Thanks to APIs, AI can connect to your e-shop, CRM or accounting — the AI fetches an order or invoice by itself and works with it. Without an API, connecting systems is harder, but there's almost always a way. Automation connected to your systems →
Chatbot
A program that talks to customers in chat — on the web, WhatsApp or Messenger. Modern AI chatbots hold natural conversations in your customers' language and answer from your data. From 2 August 2026 the AI Act requires them to be labelled as AI. AI chatbots from €239 →
Context window
The amount of text an AI can "hold in its head" at once — today's models manage hundreds of pages. The bigger the window, the more material the AI can consider when answering: a whole contract, a whole project, the entire conversation history.
Deepfake
AI content that realistically imitates real people, places or events — photos, videos, voice. From 2 August 2026 the AI Act requires labelling, which also applies to marketing: a realistic-looking AI avatar in an ad must disclose it's AI-generated. AI content labelling rules →
Embedding
Converting text into a series of numbers that capture its meaning. Embeddings let AI find passages in documents related to a query even when they don't share the same words — the foundation of search in knowledge systems (RAG). How RAG systems work →
Fine-tuning
Additional training of a finished AI model on your own data so it handles a specific task or style better. For most business uses, though, RAG or a well-written prompt is simpler and cheaper — fine-tuning pays off only at large volumes and for specific tasks.
Generative AI
AI that creates new content — text, images, video, music, code. This includes ChatGPT, Claude, Midjourney and product photo tools. Its counterpart is analytical AI, which "only" evaluates data (predictions, classification, recommendations). AI product photos & videos →
Hallucination
When AI makes up an answer that sounds credible but isn't true. The biggest practical risk of language models. It's minimised by good solution design — e.g. a RAG system where the AI answers only from your documents and cites its source. RAG with source citations →
High-risk AI system
An AI Act category for systems that make decisions about people: recruitment and CV screening, credit scoring, educational assessment, critical infrastructure. They face the strictest rules — documentation, human oversight, testing. Obligations apply from December 2027. What's postponed to 2027 →
Knowledge system
Company AI that answers questions from internal documents — contracts, manuals, policies. An employee asks in plain language and gets an answer with a link to the source, instead of digging through folders. Technically built on RAG. A knowledge system for your company →
Language model (LLM)
AI trained on vast amounts of text that understands and produces language — the engine behind ChatGPT, Claude and Gemini. LLM stands for "large language model". For a business, the LLM is the motor around which a specific solution gets built.
Machine learning
An approach where a program learns from data instead of hand-coded rules. Machine learning powers predictive models (demand or cash flow forecasts), recommendation systems and language models themselves. Predictive models & analytics →
Multimodal AI
AI that works with multiple input types at once — text, images, audio and video. In practice: send a photo of an invoice and the AI extracts the data; upload a call recording and you get a summary with action items.
Process automation
Replacing repetitive manual work with software: copying data between systems, sending notifications, processing invoices, onboarding. With AI, automation also handles unstructured inputs — like reading an e-mail and deciding what to do with it. AI automation from €199 →
Prompt
The instruction you give AI — telling ChatGPT, Claude or another model what to do. Prompt quality determines result quality more than the model itself: a specific instruction with context and an example beats a vague question every time. What a prompt is and 5 rules for writing one →
Prompt engineering
The skill of writing AI instructions so the results are reliable and usable: structuring the task, providing context, showing examples, defining the output format. Often the cheapest way for a business to get significantly more out of AI. Prompt-writing basics →
RAG (Retrieval-Augmented Generation)
A technique where the AI first retrieves relevant passages from your documents and answers only from them — with source citations. It solves the two main problems of language models: hallucinations and ignorance of your internal data. The foundation of company knowledge systems. RAG knowledge systems from €399 →
Token
The unit in which AI reads and bills text — roughly a piece of a word (a typical word is 1–3 tokens). AI services are priced per token, so operating costs scale with the volume of text the AI processes.
Training data
The data an AI model learned from. It determines what the model knows and its blind spots — the model knows nothing that wasn't in its training data (like your internal processes). That's why company knowledge is added via RAG, not by relying on the model's "memory". How to teach AI your data →
Vibe coding
A way of programming where AI writes the code and the human just describes what they want and reviews the result. It lets non-technical people build working tools — internal dashboards, scripts, prototypes — at a fraction of the previous time and cost. How AI writes code for you →
Write to me — I'll explain it clearly and with no strings attached. The initial 30-minute consultation is free.
Free non-binding consultation