Hallucination
A hallucination is an answer from an AI model that sounds convincing but is wrong: an invented price, a non-existent product, a date based on nothing.
A language model predicts plausible text, and plausible is not the same as true. Without a source it fills gaps with what sounds likely. That is harmless in a brainstorm and harmful in a quote or a delivery date sent to a customer.
The remedy is never "better prompting" alone. It is: letting the model work with your real data (RAG, integrations), answers with a source, and a human who checks before anything goes out. That is how we build agents.
Other terms
Human-in-the-loop
Human-in-the-loop means a human has a fixed place in the process: the AI prepares, the human approves or intervenes before anything becomes final.
Prompt
A prompt is the instruction you give an AI model: the question, the context and the rules. With an agent the prompt is largely fixed in advance, so it behaves the same every time.
Fine-tuning
Fine-tuning is additional training of an existing AI model on your own examples, so it masters a specific style or task better. For most companies it is unnecessary: a good prompt plus your own data (RAG) gets further.
Want to know what this means for your business?
Tell us where you lose time or margin. We show where AI pays back fastest, in plain language.