How I Built Monydragon.com for Under $100 Using 1,500 Lines

Welcome back to another episode of "What did the developer do this time?" If you enjoyed my previous descent into madness in the post about my Database Woes, you are going to absolutely love this one. Grab a cup of coffee, sit back, and let me tell you a story about how I outsourced my entire development process to a machine that occasionally thinks it's a toaster.

If you’ve visited monydragon.com recently, you might have noticed that it looks pretty slick. It functions well, the features are snappy, and the platform holds together like a well-oiled machine. But here is the kicker, the absolute plot twist of the century: this site was built with AI and is 100% prompt-driven.

When I say "prompt-driven," I don't mean I asked an AI to write a tricky regex for me once on a Tuesday. I mean I wrote over 1,500 lines of prompts to build my current site. Yes, you read that correctly. I didn't write 1,500 lines of code; I wrote 1,500 lines of English instructions to convince a Large Language Model (LLM) to write the code for me. It has been a wildly chaotic undertaking, but against all odds, I managed to produce quality code at scale.

Here is exactly how I built this platform from the ground up, the tools I used, the methodology that kept me from throwing my computer out a window, and how I did it all for less than the cost of a nice dinner.

The Flavor of the Month: Choosing My AI Sous-Chefs

Before we get into the nitty-gritty of wrangling robots, let's talk about the tools of the trade. Personally, I absolutely love the flavor of ChatGPT and Codex. Working with them is such a nice, surprisingly fluid experience—when they aren't actively trying to sabotage you, of course.

Normally, I am a creature of comfort. I love to work in fully-featured IDEs (Integrated Development Environments) like Rider or Visual Studio. Those IDEs are the luxury sedans of the programming world. They have heated seats, built-in cup holders, and they practically drive themselves. Rider, in particular, is like a warm, heavy blanket for developers.

However, for the Monydragon project, I had to switch gears. To truly embrace the AI-driven workflow, I used Cursor and Visual Studio Code paired with the Codex plugin.

Why the Downgrade in IDE Luxury?

You might be wondering why I left the warm embrace of Visual Studio. It comes down to integration. Cursor and VS Code with Codex are essentially stripped-down street racers with massive AI engines strapped to the back. They allow you to highlight a block of code, press a shortcut, and literally talk to your editor.

Having the AI right there in the editor, reading my workspace and understanding my file structure, was the secret sauce that made this 100% prompt-driven dream a reality.

The Golden Rule of Prompting: Context is King (and Hallucinations are the Court Jesters)

The fundamental secret to building with AI is this: You need to give enough context and direction so that the AI model can properly execute the tasks.

If you just walk up to ChatGPT and say, "Build me a web platform," it is going to spit out a generic HTML template from 2004, complete with a marquee tag and a hit counter. You have to treat the AI like an incredibly brilliant intern who has absolutely zero common sense. It knows every programming language ever invented, but it doesn't know why you are building what you are building.

Wrangling the Hallucinations

While giving context is crucial, you also have to constantly be on the lookout for hallucinations. For the uninitiated, an AI "hallucination" is when the model decides it doesn't know the answer, so it confidently invents one.

It will look you dead in the digital eye and write code utilizing an NPM package that does not exist. It will invent API endpoints that are pure fiction. It will write a perfectly formatted, syntactically correct function that summons a demon instead of connecting to the database.

"The AI is a brilliant coder, but it is also a pathological liar. Trust, but verify."

Because of this, my workflow requires me to rigorously review the code and actually understand what is happening. You cannot just copy and paste blindly. If you don't understand the code the AI generated, you are going to be in a world of pain when it breaks in production. I am the senior developer; the AI is the junior developer. I still have to do the code reviews.

Bite-Sized Engineering: Don't Ask for the Moon

The biggest mistake people make when trying to code with AI is asking for too much at once. If you try to build an entire app or game in a single prompt, the AI will collapse under the weight of its own context window. It will forget what it was doing halfway through, switch from Python to JavaScript for no reason, and leave you with a completely useless mess