Vibe Coding a Web Game from Scratch — Fast, Fun, and Occasionally on Fire

#webdev #web #vibecoding

Introduction

Vibe coding is the art of building software purely from instinct and creative flow. No wireframes. No JIRA tickets. Just “trusting the Force” with an AI platform like ChatGPT or Copilot. Vibe coding is what happens when you stop asking and thinking “how do I build this properly?” and start asking “what feels right right now?”

It’s fast, addictive, and when things click, it delivers instant dopamine. But it can get messy — really messy — real quick.

Why I Vibe Coded a Whole Game

My son and I “invented” a new card game for us to play and we had a lightbulb moment. “This could potentially be a game for everyone could enjoy.” I wanted to reach a large audience without shelling out money to get it tested and decided first to create a digital prototype. Since vibe coding is the trending topic in the industry today, it was a perfect opportunity for me to see what the fuss is all about.

What followed was a streak of successful prompts — designing the UI, shaping the mechanics, fixing things on the fly. I was possessed, tweaking everything as I went, solving UX problems as they appeared in front of me.

Most of it worked. Then disaster and frustration crept in. I did manage to get to a working prototype that I am happy with, and learned a lot about what vibe coding is, isn’t and where it can take you.

Elementara: Spirit of the Dice — The Web-based Project

Elementara is a 1v1 dice-based creature battler. You control a squad of three elementals, fighting until the last creature stands. Combat is determined by dice rolls — misses, hits, criticals — and each elemental has unique strengths and resistances. I added a twist: the ‘Corrupted Dice’ mechanic, which buffs a player after they roll the same result twice in a row. It adds an element of surprise into what could’ve been a pure RNG (random number generator) slugfest.

This sounded like a complex game that would take up weeks, probably months to get a working prototype up and running. I felt that this is the perfect testing ground to see if I can build the entire project in half the time. I wanted to see if I could build it in a fraction of the time a typical dev team would, push generative AI to the limits and see what it can do and debunk the myth “will generative AI replace developers?”

The Vibe Coding Process

I was hyped up, ready to start with the project. After all, it’s my first game that I would be developing and I would be vibe coding through the entire process. The vision was clear as well, “I want my game to be easy to use by many, intuitive and fun to play with for all ages.” The tools were simple: generative AI. ChatGPT for ideation, Copilot for illustrations, and a mix of v0 and Copilot for code generation.

ChatGPT — The Game Designer

ChatGPT was my best friend from the start. I gave it specific instructions on how I could integrate a card-battle game with a dice and we managed to come up with the idea of using a 6-sided dice to be the decider of the damage, somewhat similar to Dungeons and Dragons. ChatGPT also helped me come up with the names of the elementals, their hit points, strengths and weaknesses. It also helped me polish the game mechanics and rules to make the game more balanced without sacrificing the “fun” aspect of the game.

Copilot (Image Gen) — The Art Department

Amongst all the image generative AI platforms available in the market, Copilot was the only one that managed to please me with its output. I gave ChatGPT, Copilot and Gemini the equal opportunity to create images based on my prompts and very specific art style. Copilot was the only platform that gave me images that are as close as possible to the art style I wanted for the game. It was the only platform that has provided me with a lot of generated images before hitting its limits, unlike ChatGPT which maxes out at 3–4 images.

v0 & Copilot (Code) — The Engineers

Vibe coding with AI code tools was tricky but effective — until it wasn’t.

v0 was amazing at scaffolding the app based on prompts. But because of the free tier, I had to be precise with my requests. I started by feeding it the rulebook and watching the UI and game flow emerge. It worked — mostly.

Copilot, integrated in VSCode, filled in the gaps. It helped tweak logic and connect components, especially once v0 started falling apart. v0 couldn’t determine which component I was pertaining to and ended up changing the entire UI of the game.

Key Findings & Lessons Learned

Most of my coding decisions were based on instincts. I eyeballed bugs, adjusted UI issues immediately, and kept rolling. But I had to constantly remind myself: be very specific with prompts, or AI will interpret things however it wants.

In an accumulation of roughly 40–50 hours of development, I managed to wrap up Elementara in a nice red ribbon. I was surprised and very satisfied with the results where it was 90% accurate most of the time and provided to me in a few seconds.

After a few hundred prompts and getting the game to 75% completion, I started to receive hallucinations from AI and frustrations build up quickly. The main issue that annoyed me was that v0 started to change more of the UI when I specifically told it not to. Another issue was that the platform started to “stall” and wrote large parts of the codebase arbitrarily, no matter how detailed my prompt was. Lastly, v0 could not pinpoint the main issue that I wanted to get fixed. The last 25% of the project involved digging into the code myself and fine-tuning everything manually.

As much as I enjoyed vibe coding through this side project, it is not recommended for any production work. It can help your team to get up to speed in prototyping a single component. But beyond that needs manual intervention and developer experience in fine tuning to the requirements. I have not yet explored how to extend the game or whether the code is structured properly. I am sure though that I will be opening a can of worms when I do.

Conclusion

If you’re a solo dev, indie creator, or just trying to ship something fun — vibe coding is a legit weapon. Generative AI can get you from idea to prototype faster than any team ever could.

But if you’re in production? Managing a team? Building long-term?

It’s a trap.

AI can scaffold code fast — but it still needs human judgment. Left unchecked, it breeds technical debt, unstable logic, and brittle foundations. The deeper you go, the messier it gets.

Use it. Abuse it. Just don’t trust it blindly.

Vibe coding will leave you with: