Language Model (LLM)
A Large Language Model (LLM) is an AI that has learned to understand and generate language from vast amounts of text. It forms the basis for chat assistants. On its own, it doesn't know your company. Only connecting it to your data makes it useful.
AI Agent
An AI Agent is an AI that not only responds but also acts using tools: scheduling an appointment, drafting an email, recording an invoice. It breaks down a goal into steps and executes them, within the scope of the permissions you grant it.
Prompt
A prompt is the instruction given to an AI, meaning what you write or say. Good business AI doesn't require elaborate prompts. 'Remind all customers with outstanding invoices over 14 days' is sufficient.
Hallucination
A hallucination is a response that sounds convincing but is incorrect. It occurs when a language model fills gaps with probable information. Responses based on your actual data, source citations, and human review for critical decisions help prevent this.
RAG (Retrieval-Augmented Generation)
RAG means: Before responding, the AI retrieves relevant information from your data, such as emails, documents, or customer records, and bases its answer on that. This way, it knows what applies in your business without the model needing retraining.
Autonomy and Human Oversight
Autonomy describes how independently an AI is allowed to act: only suggest, act after approval, or operate autonomously. Human oversight means a person sets this level, reviews results, and can stop the AI at any time. The EU AI Act requires this for high-risk systems.
Skill
A skill is a recurring process that an AI has learned, for example, 'Answer inquiry, prepare quote, follow up in three days.' Good systems suggest skills and only use them after human consent.
Data Processing Agreement (DPA)
An agreement under Article 28 GDPR between you and a service provider that processes personal data on your behalf. Do not input customer data into an AI system without a DPA.
EU AI Act
The EU's AI regulation. It categorizes AI systems by risk and mandates obligations such as AI competence for employees (Article 4), human oversight for high-risk systems (Article 14), and transparency towards individuals interacting with an AI (Article 50).
Where Humans Decide
You Decide
- Terms help with understanding. What matters is what a system concretely does for your business.
- Ask every provider what data they use and who has control.
