Most "prompt engineering" advice is about tricks. Specific phrases. Magic words. Structured templates.
Some of it works. Most of it misses the point.
The actual skill isn't knowing what words to use — it's developing a way of thinking that translates naturally into prompts the AI can act on.
What the AI Needs
An AI coding assistant doesn't need perfect grammar or magic syntax. It needs:
- Context: What exists already. What the codebase looks like. What you've tried.
- Intent: What you're trying to achieve, not just what action to take.
- Constraints: What it shouldn't do. What rules to follow. What to preserve.
Most bad prompts fail because they skip one of these.
The Intent vs. Action Distinction
This is the most important one.
Bad prompt: "Add a button to the form." Better prompt: "I have a contact form. When the user submits it, I want to show a success message without reloading the page. Add what's needed to make that work."
The first tells the AI to do something. The second tells it what you're trying to accomplish. The AI can figure out the action — your job is to communicate the goal.
Giving Context
The AI doesn't know your project. It only knows what you tell it in the current conversation.
Before asking for something complex:
- Share the relevant code
- Describe what the component/page/feature is supposed to do
- Mention any constraints (we're using X library, we don't want Y approach)
This isn't just being thorough. It changes the quality of the output dramatically.
Constraints Are Underused
Telling the AI what not to do is as important as telling it what to do.
"Don't change the overall layout." "Keep the existing color variables — don't hardcode hex values." "Don't add any new dependencies."
These constraints prevent the AI from making technically-correct-but-wrong choices that you then have to undo.
The Iteration Loop
A single prompt rarely gives you the final answer. The workflow is:
- Prompt with intent + context
- Review the output — what's right, what's wrong
- Prompt again with what needs to change
- Repeat
The best practitioners get to the right answer in 3-5 iterations on a complex problem. Beginners sometimes think they need to get it right in one shot. You don't.
Practice
The only way to develop the prompt mindset is to use it. Write a prompt. See what happens. Notice where it went wrong. Fix it.
Over time, you develop an intuition for what the AI needs. That intuition is the skill.