Prompt Engineering for Programmers: Getting AI to Generate Better Code

Why Prompts Matter

The same AI tool with different prompts produces vastly different results. A good prompt can transform AI-generated code from "barely functional" to "production-ready."

Golden Rules for Prompt Writing

1. Specific > Vague

❌ Bad Prompt:

Write a validation function

✅ Good Prompt:

Write a TypeScript function validateEmail with requirements:
- Parameter: email: string
- Return: { valid: boolean; error?: string }
- Validation rules:
  1. Must contain @ symbol
  2. Must have characters before and after @
  3. Domain part must contain .
- Use regex implementation
- Add JSDoc comments

2. Provide Context

Tell AI about your tech stack and constraints:

In my Next.js 14 project (using App Router),
create a user login API route with requirements:
- Use zod for request body validation
- Use bcrypt for password comparison
- Return JWT token
- Handle all possible error scenarios

3. Give Examples

Based on the following code style, create a function to get order details:

// Example: Get user info
export async function getUserById(id: string) {
  const user = await prisma.user.findUnique({
    where: { id },
    select: {
      id: true,
      name: true,
      email: true,
    }
  });

  if (!user) {
    throw new NotFoundError('User not found');
  }

  return user;
}

// Please create: Get order details, including order items and user information

4. Break Down Complex Tasks

Split complex tasks into steps:

Step 1: First help me design the database schema,
        I need a blogging system with articles, categories, tags, comments

Step 2: (After getting schema)
        Based on this schema, create Prisma models

Step 3: (After getting models)
        Create CRUD API for articles module

Practical Prompt Templates

Template 1: Code Refactoring

Refactor the following code with requirements:
1. Follow SOLID principles
2. Extract repeated logic
3. Add appropriate error handling
4. Maintain original functionality
5. Add type annotations

Original code:
[Paste code]

Template 2: Bug Fixing

This code has a bug that manifests as [describe symptoms].

Expected behavior: [describe expected]
Actual behavior: [describe actual]

Please:
1. Analyze possible causes
2. Provide fix solution
3. Explain why this fixes it

Code:
[Paste code]

Template 3: Code Review

Please review the following code from these perspectives:
- Security vulnerabilities
- Performance issues
- Code readability
- Potential bugs
- Best practices

Code:
[Paste code]

Template 4: Learning New Technology

I'm a React developer with 3 years of experience,
and now need to learn [new technology name].

Please:
1. Explain core concepts using React analogies
2. Provide a simple entry-level example
3. List 3 common pitfalls
4. Recommend learning path

Advanced Techniques

Role Setting

You are a backend architect with 10 years of experience,
focused on high-concurrency system design.

Please evaluate the scalability of the following API design...

Constraints

Complete the task under these constraints:
- Don't use any third-party libraries
- Keep code to under 50 lines
- Time complexity must be O(n)

Chain of Thought

Think through this problem step by step:
1. First analyze requirements
2. Consider possible solutions
3. Evaluate pros and cons of each
4. Choose best solution and implement

Common Mistakes

1. Too vague: No specific requirements
2. Too wordy: Too much irrelevant information
3. Missing context: Don't mention tech stack and constraints
4. Asking too much at once: Not breaking down complex tasks

Summary

Good Prompt = Specific Requirements + Sufficient Context + Clear Constraints + Expected Output Format

Spend 2 minutes writing a good prompt, save 20 minutes debugging AI-generated code.

---

Practice Tip: Create a prompt template library, save your commonly used prompts, and just tweak them each time.