Prompt Engineering

​

The Reality of AI Prompting

Most people think prompting is typing whatever comes to mind and hoping the AI figures it out. That approach wastes time. The difference between struggling with AI and having it execute your vision precisely comes down to how you communicate. Aurora uses large language models that respond to patterns in language. They don’t read minds, guess context, or fill in blanks the way humans do. They follow instructions literally. This isn’t a limitation—it’s an opportunity. Clear instructions get clear results.

​

Why This Matters

Better prompting means:

• Less iteration - Get what you want on the first or second try instead of the fifth
• Faster debugging - Point the AI to the exact problem instead of vague complaints
• More control - Direct the AI’s decisions rather than accepting whatever it generates
• Deeper capabilities - Access advanced features you didn’t know existed

You don’t need to be technical. You need to be precise.

​

How AI Actually Works

Large language models predict the next most likely tokens based on your input and their training data. Understanding this changes how you approach prompting:

​

The Model Has No Context Beyond Your Prompt

It doesn’t know your project history, your preferences, or common sense assumptions. Everything relevant needs to be in your message. Bad: “Make the login better”
Good: “Add password validation to the login form: minimum 8 characters, at least one number, at least one special character. Show error messages in red below the input field.”

​

Structure Determines Output Quality

Models pay more attention to information at the beginning and end of your prompt. Put critical requirements first. Long prompts can cause the model to lose track of earlier instructions—stay focused.

​

The Model Will Sound Confident Even When Wrong

It generates plausible-sounding text regardless of accuracy. If you need factual information, provide reference materials or verify the output yourself. For code, always test.

​

Be Literal and Explicit

Think of it like instructing someone who follows directions exactly but makes no assumptions. If you don’t specify something, don’t expect it to appear in the output.

​

The CLEAR Framework

Use these principles to evaluate and improve your prompts: Concise
Remove filler. Every word should serve a purpose. “Write a 200-word summary of climate change effects on coastal cities” beats “Could you maybe write something about climate and coasts?” Logical
Break complex requests into ordered steps. Instead of “Build a signup feature and show usage stats,” separate them: “First, implement user signup with email/password via Supabase. Then create a dashboard displaying user count statistics.” Explicit
State exactly what you want and don’t want. Provide format examples. “List 5 unique facts about Golden Retrievers in bullet points” is better than “Tell me about dogs.” Adaptive
Iterate. If the first output misses, clarify in a follow-up message: “The solution is missing authentication. Add user auth with Supabase.” Refine until you get what you need. Reflective
Note what works. After successful prompts, review what made them effective. Build a mental library of patterns that get good results for your use cases.

​

Structured Prompting Format

For complex tasks, use labeled sections to organize your requirements: Context
Background information the AI needs. Example: “This is a multi-tenant SaaS application where users belong to organizations.” Task
The specific goal. Example: “Create an organization switcher in the navbar that lets users change between their organizations.” Guidelines
Preferred approach or tools. Example: “Use a dropdown menu styled with Tailwind. Store the selected organization ID in localStorage.” Constraints
Hard requirements or restrictions. Example: “Users must stay on the same page when switching organizations. Don’t refresh the entire app.” This structure forces you to think through requirements before sending the prompt. It also gives the AI clear boundaries to work within.

​

Common Patterns That Work

​

Building Features

Template: “Add [feature] to [location] that [behavior]. Use [technology/approach]. [Additional constraints].” Example: “Add a search bar to the top of the products page that filters results in real-time as the user types. Use Supabase full-text search. Display a loading indicator while searching.”

​

Fixing Bugs

Template: “The [component] currently [incorrect behavior]. It should [correct behavior]. [Relevant context about why it’s happening].” Example: “The form submission currently refreshes the page. It should submit via fetch API and stay on the same page. The form element has onSubmit but preventDefault isn’t being called.”

​

Refactoring Code

Template: “Refactor [component/function] to [improvement goal]. Keep [must preserve]. Change [what should change].” Example: “Refactor the UserProfile component to use React Query for data fetching. Keep the existing UI exactly the same. Change only the data loading logic.”

​

Styling Changes

Template: “Update [element] styling to [visual goal]. Use [design system/colors]. Match [reference].” Example: “Update the call-to-action button to have a gradient background from purple-600 to indigo-600. Use Tailwind classes. Match the hero section styling.”

​

What Not To Do

Vague requests - “Make it look better” gives the AI nothing to work with. Assuming context - “Fix the error” when there are multiple errors leaves the AI guessing. Overly complex single prompts - Break large features into sequential smaller requests. Ignoring errors - If the AI’s output has bugs, describe them specifically rather than just saying “it doesn’t work.” Expecting perfection immediately - Plan for 2-3 iterations on complex features.

​

Advanced: Meta Prompting

Once comfortable with basics, you can ask the AI to improve your prompts: “I want to add user authentication to my app. How should I structure my prompt to get the best implementation from you?” The AI can suggest what details to include, what order to present information, and what potential issues to address upfront.

​

Next Steps

Start applying CLEAR principles to your Aurora prompts. When you get a poor result, review which principle was violated. When you get a great result, note what made the prompt effective. The goal isn’t perfect prompts every time—it’s continuous improvement in how you communicate with AI. Each interaction teaches you more about what works for your specific projects and style.