Software that interprets a question asked in ordinary language and produces an answer. Modern ones use language models rather than a decision tree of scripted replies.
The model supplies the language. Where the facts come from is a separate decision, and it is the one that determines whether the thing is useful or dangerous.
A scripted bot follows a flow somebody designed: reliable, predictable, and unable to handle anything unanticipated. A generative one interprets freely and can answer things nobody scripted.
The trade is control against coverage. Generative bots fail differently — not by saying "I did not understand" but by producing a confident, plausible, wrong answer.
It will, unless it is restricted to answering from retrieved records and required to cite them.
No. Read and draft; anything that changes a record or moves money goes to a person.
Generative with grounding, for most support use. Scripted flows still suit narrow, high-stakes paths.
Half an hour with your own data usually saves reading three of these. The guides will still be here afterwards.