DEV Community

Kade Anders
Kade Anders

Posted on

I Was Frustrated with Bad AI Code, So I Built a "Prompt Architect" to Fix It

I love AI coding assistants. Tools like GitHub Copilot and the chat inside Cursor feel like magic. They've fundamentally changed how I build software. But for me, that magic came with a frustrating catch.

I found myself stuck in a painful loop:

  1. Have a cool idea for a feature.
  2. Write a quick prompt describing it.
  3. Get back code that was almost right, but buggy or incomplete.
  4. Spend the next 20 minutes tweaking the prompt, fixing the AI's mistakes, and re-explaining myself.

The problem wasn't the AI. The problem was my prompt.

I was giving the AI a vague sketch and expecting a finished blueprint in return. I'd forget to specify the language, the error handling, the edge cases, or the exact data structures. The AI, in turn, would have to guess, and its guesses were often just slightly off.

I realized I didn't need a better AI. I needed a better process. I needed something that would act like a seasoned product manager or a senior developer—a partner that would interview me, poke holes in my idea, and force me to think through the details before a single line of code was generated.

Introducing Clarion: Your AI Prompt Architect

Since that tool didn't exist, I decided to build it. I call it Clarion.

It's a simple, free web app designed to do one thing: turn a vague idea into a flawless, comprehensive specification for an AI developer.

You can try it live right here:
Clarion

The "Ruthless Architect" Approach

The core idea behind Clarion is to be a professional tool, not a friendly chatbot.

Instead of trying to guess what you mean, Clarion's AI brain analyzes your goal and the conversation history. If it detects a missing piece of critical information (like the programming language, the required inputs, or the desired output), it stops and asks you a single, direct question.

You say: "I need a script to process a CSV file."
Clarion asks: "What programming language should be used?"
You say: "Python."
Clarion asks: "How should the final result be presented? Printed to the console or saved to a new file?"

It works one question at a time to reduce cognitive load. This back-and-forth continues until it has the absolute minimum information required to build a robust plan.

Once it's satisfied, it generates a detailed, structured Markdown specification. This isn't just a prompt; it's a professional-grade technical brief that you can copy and paste directly into Cursor, ChatGPT, Claude, or any other AI coding assistant.

How It's Built (The Tech Story)

I wanted to build Clarion with the same lean, efficient philosophy as the app itself.

Frontend: Built with Next.js and Tailwind CSS.
Backend: A single Next.js API route deployed on Vercel.
The Brain: The AI is powered by a variety of models available through the GitHub Marketplace, including GPT-4o, Grok, and DeepSeek.

The entire project was built and deployed with a $0 budget.

Give It a Try

I'm launching Clarion today because I'm genuinely curious if other developers face the same frustrations I did. I'd love for you to try it out and let me know what you think.

Does it help you clarify your own ideas? Is the final prompt it generates actually useful? What's missing?

I'll be hanging out in the comments to answer any questions. Thanks for reading

Top comments (0)