This section asks three questions: what AI can do, why it fails, and what people remain responsible for.
A useful starting framework
- A model is not a reference database: it generates from context and does not automatically possess correct facts.
- A prompt is not a magic spell: a clear goal, source material, boundaries, and acceptance criteria matter more than stacked tricks.
- Tool capability is not task completion: a successful call is process evidence; the result still needs verification.
- Automation needs responsibility boundaries: the closer an action gets to publishing, payment, deletion, or account access, the clearer human confirmation must be.
Current notes
Return to the knowledge map.