Chapter 02 · ~12 min

How to talk to AI.

Prompting is a real skill. Most people don't have it. By the end of this chapter, you'll be in the top 10% of users.

TL;DR: the four moves

  • Prompting is the new writing. Same skill, new shape.
  • Use four ingredients: context, role, examples, constraints.
  • First answer = draft. The conversation is the product.
  • Don't outsource what you should be learning. Make yourself sharper.

Prompting is the new writing

For a thousand years, "good writing" meant arranging words so a human understood you. Now there's a second kind: arranging words so an AI understands you.

Same vocabulary. Different moves. This is the skill the next decade runs on: and almost no one is teaching it.

Bad prompt vs. great prompt

Same goal, two ways: "help me understand photosynthesis."

Bad

explain photosynthesis

Generic textbook paragraph. Forgettable.

Great

I'm a 9th grader studying for a bio
test Friday. I know plants make food
from sunlight but not the mechanism.

Teach me. Use one analogy I can
picture. Then ask me 3 questions to
test if I got it.

Tailored explanation, vivid analogy, active quiz.

The great prompt isn't long because it's flashier. It's long because it carries information the model needed to do a good job.

The four ingredients of a great prompt

You don't always need all four. But you'll be amazed how often you should.

01

Context

Who you are, what you're doing, what level you're at.

02

Role

Tell the AI who to be. "Act as a strict editor." Highest leverage move there is.

03

Examples

Show, don't tell. Paste in 2-3 examples of what good looks like.

04

Constraints

"In 100 words." "3 bullets." "For a 14-year-old." Force a shape.

Each one, with a worked example

Role: the cheat code

Act as a brutally honest writing coach.
Read this paragraph and tell me what's
not working. Be specific.

Roles shift vocabulary, tone, and what gets emphasized. Try the same prompt without a role: completely different answer.

Examples: show, don't tell

Help me name a new podcast. Style:

- "Reply All"
- "Hard Fork"
- "Search Engine"

Short. Two words. Slightly mysterious.
Generate 10.

Called few-shot prompting. Two examples can change everything.

The cardinal rule: don't accept the first answer

Beginners send a prompt, read the reply, move on. Pros treat the first reply as a draft.

  • "Make this shorter."
  • "Now argue against your own answer."
  • "What did you miss?"
  • "Where could you be wrong?"

The conversation is the product, not the first message.

The biggest beginner mistake

Asking AI to do things you should be learning. "Write my essay." "Solve my problem set."

The AI will do it. The cost is your own development. Research now shows that people who outsource thinking before they've built the underlying skill end up worse at the skill.

The better move

Use AI to learn the thing faster, not skip past it. "Teach me X." "Test me on Y." "Help me debug my thinking on Z." That partnership makes you sharper. Outsourcing makes you weaker.

Try it

Send this prompt to a real model. Notice how role + constraint produces something five words couldn't.

Open playground