Build a client quote list: client name, amount, date, and status, stored in the browser, newest first.
A prompt is the message you send the AI—the problem, requirements, and context all live in it, and it steers what the AI builds.·"Build a client quote list with client, amount, date, and status, stored locally, newest first"—that whole sentence is the prompt. The AI reads it before every move: specific wording lands specific results, while "make a nice tool" forces guessing. When results disappoint, refine the prompt before switching tools—add the object, the requirements, and what counts as done.
System prompt · preset, invisible to youYou are a frontend assistant; touch only the files the user names…
Your prompt · what you write each turnBuild a quote list: client, amount, date, status; store locally; newest first.
Output · grows from your promptQuote list: full CRUD, four fields, newest first ✓
Off-target result? Fix the layer you writeAdd the object, requirements, and done-criteria—far faster than switching tools.
Three things that make a prompt concrete
Say what you want—and what you don't: Exclusions like "no charts yet" or "no login" stop the AI from growing the tool on its own.
Give checkable done-criteria: "Works on mobile" and "add, edit, delete all work" give you and the AI the same ruler.
One round, one change: Changing many things at once hides which sentence caused the error. Change one, verify, then continue.
You can say this to an AI Agent
“
Build a client quote list with client name, amount, date, and status, stored in the browser for now. Skip charts and export this round. Done means: add, edit, and delete all work, and it works on mobile too.
