You don't need a class for this — being specific about what you want, in the message itself, is basically the whole skill. Watch the same request get a genuinely better answer as you add detail, one piece at a time.
The scenario: You run a small online shop. A customer just emailed: "This is the SECOND time my order arrived late. I want a refund." You ask an AI to help you draft a reply.
1. The vague version — what most people actually type
This is the entire prompt: "Write a reply to this email."
You typed
Write a reply to this email.
AI's draft
Thank you for reaching out. We apologize for any inconvenience caused. We are looking into this matter and will follow up soon.
Technically an answer. Also completely useless — it doesn't even mention a refund, a delay, or say sorry for something specific.
2. Add detail to the SAME message — nothing saved, nothing remembered
This is prompt engineering: adding specifics to what you type, right now, in this one message. Flip these on — the draft below updates instantly.
What you add to your message
You typed
Write a reply to this email
AI's draft
Nothing here calls a real AI model — it's a scripted simulation so the effect of each addition is instant and obvious. The real thing works the same way: more specific instructions in the message itself, better-matched output.
3. Now predict it yourself no peeking
You've added all four prompt techniques above and gotten a great reply. A friend asks: "Does the AI remember to do that automatically next time you open it?" What actually happens if you close this chat and start a brand new one?
Yes — once you've shown it a good example, it remembers to keep doing that.
No — a new conversation has nothing carried over. You'd have to type all four again.
It depends on how many times you've asked something similar before.
That's the catch this whole lesson is building to. Prompt engineering only ever affects the message you're in right now. That limitation is exactly the problem the next lesson (context engineering) exists to solve: writing it down once, instead of retyping it every single time.
The part that trips people up
This only affects this one message. Close the chat, open a new one, and the AI has no memory that you ever said "don't promise a refund yet" — you'd have to type it again. That's fine for a one-off question. It gets old fast when it's the same handful of rules, every single time, for every customer email you ever draft.
The next lesson solves exactly this: write your rules down once, in a file the AI reads automatically, rather than retyping them into every message. That's called context engineering.