The repair café invitation brought the whole module together. The app supplied the working information. Your request defined the job. You checked what the draft kept, omitted, and invented.
That distinction also explains why assistants behave differently. One app may supply saved notes or fetch an event page. Another may offer the same model neither.
A refusal can arise from training, app instructions, or a separate filter. A current answer may rely on supplied information or a search. You cannot identify the cause from the writing style alone.
The next module follows that surrounding software. When the model requests a search, a calculation, or an action, what actually carries it out?
The whole chapter, simply
Give the model the information it needs and a clear job to do.
Check the result against evidence, with stronger checks when mistakes matter.
The model generates output; the surrounding application supplies context and tools.