Writing a Good Prompt

Prompting is the new coding. The logic is the same: garbage in, garbage out. If you give an LLM a vague instruction, don't be surprised when you get a generic result.

Context is king. Treat the AI like a brilliant junior developer who doesn't know your codebase or your preferences.

The Bad Way

This is what most people do, and why they think AI is "meh".

Write code for a button.

Result: You'll get a generic HTML button, or maybe a React component if you're lucky. No styling, no props, no context.

The Good Way

Be specific. Define the constraints, the tech stack, and the edge cases.

Create a reusable Button component for a Next.js 14 app using Tailwind CSS and TypeScript.

Requirements:
- Variants: 'default', 'destructive', 'outline', 'secondary', 'ghost', 'link'.
- Sizes: 'default', 'sm', 'lg', 'icon'.
- Props: Extend React.ButtonHTMLAttributes<HTMLButtonElement>.
- State: Add an 'isLoading' prop that disables the button and shows a spinner icon (lucide-react).
- Usage: It should be accessible and use forwardRef.

Style it to match a clean, minimal aesthetic (shadcn/ui style).

Result: You get a production-ready component that you can drop straight into your project.

The Mental Model

Think of a prompt as a function definition.

• Parameters: What inputs are you providing? (Code snippets, error logs).
• Body: What transformation do you want? (Fix, refactor, explain).
• Return Type: What format do you want? (JSON, Markdown, Python code).

Stop guessing. Start specifying.