This article is the third entry in the series, following Making a Game with Vibe Coding: Part 1 and Part 2.
There was good news.
'Twilight Mountain Defense,' a game I created, won the Grand Prize at our 'In-House AI Game Development Contest 2026.' Since I happened to be serializing articles under the theme of 'Making a Game with Vibe Coding,' this contest brought an unusually heavy burden.
Although I had never learned game development or coded before, I felt I had put in enough hard work to really want an award. The result turned out far better than expected, and to be honest, I was thrilled.
However, the joy was short-lived.
"Congratulations. Also, an AI game development event is taking place in August called 'OpenAI Game Builders Seoul.' Why don't you enter it with that game?"
"Pardon?"
"I'll give you work hours for it and support you with AI tools."
"...Got it."
And just like that, game development, which I thought was finished, began anew.
Deciding to Remake the Award-Winning Game from Scratch
I was flustered, but I didn't hate the idea.
Making the first game was genuinely fun. Iterating through conversations with AI, seeing the project gradually take shape, and realizing the things I had imagined one by one completely absorbed me.
So when given the opportunity to make a game again with guaranteed work hours, I had no reason to refuse.
The problem was time. There were only about two weeks left until submission. Looking at the competition criteria, simply tweaking an existing game didn't seem like enough.
The event required submitting a game prototype developed with Codex along with the development experience. While using an existing project was allowed, 'Twilight Mountain Defense' was primarily made using Claude. I decided to use this opportunity to rebuild it from scratch using Codex.
Since I was rebuilding it anyway, I decided to try something new. Instead of creating a single HTML file, I chose to use a proper game engine. I selected the Godot Engine, which has been frequently discussed among indie developers recently.
My interest in Godot came from an interview I conducted in the past with a solo developer. Having left Nexon to develop games independently, he praised Godot as quite a solid engine for indie game creation. The thought of trying it out someday had stuck with me ever since.
Of course, I don't know how to use a game engine (laughs). So it was only possible if AI 'did it for me.' Fortunately, Codex replied that it could create a web game using Godot Engine.
Now, there were two goals to achieve over the next two weeks.
First, remake 'Twilight Mountain Defense' from scratch using Codex.
Second, complete a browser-executable game using Godot Engine for the first time.
And so, the second game development project began.
Coding and Image Generation in One Place… Why Codex Was Great
Working with Codex was far better than expected—to the point where I found myself thinking, "It can even do this?"
First, making a game with Godot Engine. I didn't even know the difference between nodes and scenes. After downloading the Godot Engine, I provided Codex with the installation directory and project path. Codex then ran the engine directly to carry out the work. All I did was download the engine, yet before I knew it, I was building a game with a proper engine. Haha.
Another great thing was the seamless workflow. When making the original, I entrusted the code and game systems to Claude while generating graphics with a separate AI like Grok. Switching back and forth between two AIs was extremely tedious. Downloading images, removing backgrounds, and adjusting sizes were all manual labor. I had to feed those images back to Claude for feedback and repeat the production process multiple times.
With Codex, everything from character design to background removal, resizing, applying graphics in-game, and modifying code flowed seamlessly within a single workflow. When I requested, "Make the Pyromancer throw flames and the Cryomancer spread ice," it modified not only the visuals but also the attack effects and damage mechanics altogether.
There was also a huge difference in graphical work.
Previously, I had to manually remove the backgrounds from AI-generated images. Lacking professional image-editing skills, background colors often remained along character outlines, or parts of hair and weapons got cut off. Zooming in on images inside the game made those flawed crops stand out even more clearly.
This time, I left image creation, post-processing, and in-game application all to Codex. Comparing character proportions and sizes, scaling up small units and shrinking large ones were all done based on the actual gameplay screen.
As a result, character outlines became much cleaner, and even across different tiers and classes, the art style gained a cohesive look as if from a single game.
It wasn't just the characters that changed. I also revamped the moments players remember most vividly, such as obtaining high-tier units or boss entry scenes.
Codex was better than the previous setup in almost every aspect, but it had weak spots. What I disliked most was the BGM.
"Keep the melody of the original and arrange it like a video game orchestral score."
A moment later, the music came out—sounding like a ghost wailing.
"Not this. Try again."
No matter how many times I regenerated it, the music barely improved. Part of it was that the original's cheap arcade BGM vibe had turned out surprisingly well, but it made me realize that crafting and selecting memorable, pleasing melodies remains a domain where human intuition is crucial. After all, choosing what sounds good to human ears is fundamentally a human task.
Lastly, I must mention Codex's debugging capability. Because it tested execution and errors after every task before delivering results, instances where I had to find bugs and request fixes decreased drastically compared to before. The generous usage limit was another feature strong enough to make me fall for Codex.
The Completed Game… Codex Calculated a 70% Chance of Reaching the Finals
Through over 70 revisions big and small, the game was finally completed.
Ahead of submission, I asked Codex:
"What are the chances of passing the first preliminary round with this game?"
Codex evaluated the game in considerable detail.
It cited as strengths that the game runs directly in a browser and offers clear core fun through summoning, combining, acquiring Mythic-tier units, and clutch situations. It analyzed that compared to the original, characters, monsters, attack effects, UI, and boss staging had significantly improved, and that actual gameplay footage alongside the story of 'a journalist with no coding background remaking a game with Codex' would catch the judges' attention.
There were weaknesses too. As random defense is a familiar genre, its first impression might lack originality; late-game screen clutter from units, monsters, and effects pouring in at once; and stability/optimization being wildcards due to it being a web build made in a short time frame.
Synthesizing these factors, Codex's prediction was as follows:
It didn't just throw out numbers. It explained why there was a chance of passing, where points might be deducted, and what needed further refinement right before submission. The analysis was so detailed that it raised my expectations.
However, there was a major assumption hidden behind that 70%. With limited public information available, Codex referenced hackathons of similar scale and assumed around 80 participating teams. Since this competition selected 40 teams for the finals, it simply assumed about half the applicants would make it through.
Eagerly Awaited Results End in Elimination… The Competition Was Far Tougher
After waiting through the entire weekend, the result I received was elimination.
It was disappointing. I had spent a long time remaking the game and felt confident it was better than the original. Right after checking the result, I wondered if random defense being a common genre and weaknesses in originality had held me back.
Of course, there is no way to know the exact reason for elimination. The evaluation criteria disclosed by the organizers consisted of five areas: playability, originality, collaboration with Codex, marketability/releasability, and presentation. Rather than a single metric, the selection was likely based on a comprehensive evaluation of all factors.
I only learned after the rejection was confirmed, but the scale of participation was far bigger than expected. According to an OpenAI representative's social media post, over 1,000 teams applied for 'OpenAI Game Builders Seoul.' Based on that figure of over 1,000 teams, the acceptance rate for the 40 finalist slots was below 4%, with a competition ratio exceeding 25 to 1. It was an entirely different scale from my initial assumption of 40 out of 80 teams.
Furthermore, many active professional game developers reportedly teamed up to enter. Hearing that made me realize how naive I was to participate alone and expect to reach the finals.
Still, I'm Glad I Remade It
While the result was disappointing, the process was enjoyable.
It was a truly satisfying challenge in that I significantly advanced a game I made with passion. I completed a web game using Godot Engine for the first time, remade every character and monster from scratch, and implemented combinations, augments, boss staging, automated testing, and even an EX Endless Mode. While creating the game, I was completely absorbed without worrying about anything else. To exaggerate slightly, I felt that making a game was even more fun than playing one.
It would have been even better to reach the finals, but you don't always get the outcome you hope for. Now that I've discovered the joy of making games, I feel like I could challenge myself with a different genre next time.
The vibe coding game development series written over three installments has now reached its conclusion. It took a long time and resulted in only three articles. I still feel a bit of regret, wishing I had been a little more diligent or written more entertainingly.
Still, I am happy to conclude this series alongside a tangible project. If given the opportunity to develop a game through vibe coding again, I would gladly take on the challenge. Thank you to everyone who followed along with this journey.
