How I Won Silver at the First Vibe-Coding1 Grand Prix

Introduction

I was fortunate enough to receive a silver medal at the recently held First V-1 Grand Prix (V-1).

The organizer was Mr. Anno, who has garnered attention for running for governor of Tokyo as an AI engineer. I am truly honored to have won a medal among such distinguished participants.

However, this was a pilot version with only six participants. So, while the silver medal isn't a huge accomplishment, it's still a medal, and I'd like to share my expertise as a vibe-coding expert in this article.

This article will cover two key points:

1. What the First V-1 Grand Prix was like.

2. My vibe-coding approach that led to my silver medal win.

I hope this is helpful and enjoyable for you all.

Game Created: 10-Second City

The First V-1 Grand Prix

The First V-1 Grand Prix began with Mr. Anno's post.

I happened to see this post, thought it looked interesting, and decided to participate. I found two strong contenders focused on X Real, confirming my initial hunch.

Timetable

• Rule Setting: 20:00-20:15

• Competition: 20:15-21:15 (1 hour for the first round)

• Judging: 21:15-21:45

• Knowledge Sharing: 21:45-22:00

During the rule-setting phase, it was decided that games would be created and each person (including the audience) would cast two votes. At this point, I judged that the competition would be about fun, not just technical skill.

Creating a technically impressive game in 1 hour is nearly impossible. In Vibe-Coding, you'll encounter unfamiliar code, increasing the risk of unforeseen bugs that eat up your time.

Therefore, I decided to focus on an idea and UI that could be implemented within 10 minutes, prioritizing user engagement to win.

My Vibe-Coding Approach

When creating something quickly, I follow these three stages:

1. Idea Generation (approximately 10 minutes)

Model Used: At the time, I considered o3 my most reliable resource and used it for brainstorming. The prompt was: "Game ideas that can be played in a browser, created in about an hour, and suitable for a contest."

After several discussions on ChatGPT, the following ideas emerged:

They all seemed interesting, but I chose 10-Second City.

Here's why:

1. Zero external API dependencies

2. Emojis make the screen visually appealing

3. I like SimCity

2. Implementation Phase (approximately 10 minutes)

First, I created the repository. I find it faster to type commands and set up the environment manually. Concurrently, I had o3 generate code, pasting the output into cursor. At this point, the core game mechanics were functional. o3's ability to produce instantly executable code is truly amazing.

After several playthroughs, I found several shortcomings: the scoring system was vaguely defined and relied too much on chance; the buttons and scores were hard to see; the UI was broken, etc.

I deployed to Vercel, and then used cursor and collaborated with sonnet3.7 (who's strong in UI) to refine it.

3. Refinement (remaining 40 minutes)

Several unexpected issues arose:

• cursor kept editing non-existent files, resulting in repeated errors

• Build errors on Vercel

• The UI remained broken

• Started using a non-existent API from a non-existent library

• The cursor API failed

With vibe-coding, the real work usually starts once you have a mostly functional product.

TIP: Limiting yourself to commonly used libraries you're familiar with reduces endless debugging.

Results Announcement

• 🥇 Gold Medal: Aoyama-san — A game where emojis related to keywords are displayed, and players race to guess the keywords.

  

• 🥈 Silver Medal: Me — 10-Second City

  

• 🥉 Bronze Medal: Anno-san — Quantum Tetris (falling blocks in a quantum superposition)

  https://t.co/qlk7CD2SNP

I was surprised that everyone managed to create playable games in just one hour. With today's tools, I feel like a team could create a game ready for sale in a day.

Personally, I found Anno-san's Quantum Tetris extremely interesting (though it was quite difficult).

I'd also like to give a special award for "Most Vibes" to Nishio-san. From the start, Nishio-san used conversation to instruct Devin to create and deploy the repository, and I felt an incredible vibe, which continued throughout the competition.

Lessons Learned

1. Relying on vibes for scoring logic results in a vibe-game full of loopholes.

2. You can't do vibe-coding without an "anything-goes" vibe.

3. Relying on vibes for library selection leads to a debugging hell of outdated libraries.

Summary for Developers

After doing vibe-coding for a while, I've learned that it's crucial to understand what AI is good at and what it's not good at.

For example, in this competition:

What AI is not good at:

1. Fixing broken UI

2. Coming up with interesting rules

What AI is good at:

1. Creating a foundation

2. Correcting overall logic

Understanding these key points will prevent you from wasting time by making the AI do things it isn't capable of.

Conclusion

The V-1 Grand Prix proved that you can create a fun game in just one hour.

Next time, I'm aiming for gold with even higher vibes.

Introducing the Evame Global Developer Community

This website is Evame, a global developer community I'm building.

When developers post projects they've worked on or articles, AI automatically translates them into multiple languages.

It's still under development, so there are bugs, but please give it a try!