Start here

Begin with a job, not a tool

Open a model only when you can finish this sentence: I want help with _ so that _.

If you cannot finish it, you are browsing. Curiosity is fine. Work needs a job.

Tools multiply options. Jobs narrow them. Start narrow.

Three starter jobs (pick one today)

Use these as training wheels. Copy the job line into your first message, then add your real context.

1. Draft to edit - I want a rough outline for ____ so that I can rewrite it in my voice before anyone else sees it. Good for emails, status notes, meeting agendas. Bad for final customer wording on day one.
2. Critique with teeth - I want a skeptical review of this draft so that I find weak claims before a client does. Paste your draft. Ask for assumptions, missing risks, and what a careful reader would challenge.
3. Explain then check - I want a plain-language explanation of ____ so that I can verify it against a source I trust. Use for unfamiliar terms, error messages, or a policy you must summarize. Do not treat the chat as the source of record.

If none of these fit, write your own job line before you open the tab. If you still cannot, close the tab.

A simple first loop

1. Ask for a draft, outline, or critique - not a final truth.
2. Check anything that would cost money, reputation, or safety if wrong.
3. Edit in your own voice before you ship.
4. Save the prompt and the result that worked, so you can reuse the pattern.

That loop is enough for week one. You do not need a second tool, a prompt marketplace, or a perfect system prompt.

What models are good for today

• First drafts when you already know the destination
• Restructuring messy notes into a checklist or outline
• Generating options you will discard most of
• Explaining code or docs you can open beside the chat
• Role-playing a skeptic against your draft

What they are not

• A source of truth for law, medicine, finance, or “company policy”
• A safe place for passwords, PHI, customer dumps, or production secrets
• A substitute for running the code, opening the citation, or owning the send button
• A reason to skip the AI craft checklist

When to stop

Stop the session when any of these is true:

• You cannot name the next human check
• You are pasting data you would not email to a stranger
• You are arguing with the model instead of verifying outside it
• You have a usable draft and are polishing for dopamine, not for the reader

Enthusiasm is allowed. Endless tabs are not a skill.

Enthusiasm, timed

It is reasonable to feel excited. Models can accelerate writing, coding, research, and teaching. Keep a short list of wins so the hype does not replace memory of what actually helped you.

Next in your first hour

Go to Ask better and run one starter job with a structured prompt. Keep Verify open for the check before you paste or send. Pin the principles checklist beside your chat.