AI Chaos & Experiment:
- When I started building an engine with an AI assistant, I thought - it would be cool, fast, convenient.
- But it turned out to be a constant experiment where I never know what will happen tomorrow.
- Sometimes AI generates brilliant ideas, sometimes it goes off the rails, and I sit there thinking: can you even build something truly usable with it?
- Sometimes it feels like I'm not coding, but managing the chaos my own AI partner throws at me.
- This article is an honest experiment diary.
- No guarantee of success, but full of vibe and surprises. 🎲
Now - how I got here, and why AI is even involved... 🤖
The Birth of an Idea: "Make a Simulator of Myself? Why Not!" 💡
It all started with a sudden thought: "Why not make a programmer life simulator using a new approach - Vibecoding?" 💭
I wrote the first lines of code in seconds, and suddenly the game was almost ready.
The first "player" appeared, his sprite - and the outlines of a future virtual world started to take shape.
But that was just the beginning of my journey into the world of "vibecoding".
The idea seemed so original and meta-cool that the decision came by itself: for a game created in such an unusual way, you need your own unique engine.
Modern engines aren't suitable for working with AI assistants: their internals are closed, and AI can't fully control or change anything inside.
So I wanted to try making something of my own - open for experiments and true AI collaboration. 🛠
Vibecoding in Action: When AI Is Your Co-Author (Sometimes Very Stubborn) 🤝
An essential part of "vibecoding" is its unpredictability.
Sometimes, the AI assistant enthusiastically generates hundreds of lines of code for a new feature, but... it just doesn't work.
At all.
So I sit there, analyzing this monumental labor of artificial intelligence, trying to find the bug.
Then, after long reflection (and maybe more than one cup of coffee ☕), I spot some silly detail: a variable declared in the wrong scope, or one line of code conflicting with another.
I move that cursed variable to the right scope - and, miracle, everything magically comes to life!
In those moments, you feel like both a detective 🕵 and a bit of a wizard 🧙, realizing that even the most advanced AI can't fully replace human intuition and experience.
But this iterative debugging and problem-solving process together is the real "vibe". ✨
Sometimes arguing with AI becomes a sport of its own: it gets lazy, acts dumb, you want to curse at it, and some models (I think it was gemeni) can even sass you back.
At times like these, you realize working with AI isn't just about code - it's about nerves, patience, and a sense of humor. 😅
What annoys me most is when the model does something you didn't expect, you point it out, and it replies: "I did as you asked" and even quotes your own words.
That's when you really want to turn off the computer and go drink vodka. 🍵
I work in parallel with all the latest models: sonnet 3.7, gemeni 2.5, gpt 4.1.
And honestly, only sonnet and gemeni let you actually write code and get sensible answers - the rest either do nothing or try to build a parallel architecture for a tiny feature. 🤦
Sometimes you spend hours trying to "persuade" AI to generate something that works, editing, regenerating, editing again...
And at some point you realize it's more efficient to just hit "reset", delete all the AI's creative output, and start over.
Or, with a deep sigh, just sit down and write that piece of code yourself, old-school.
That's also "vibecoding" - mental flexibility and knowing when to trust the machine, and when to trust your own hands and head. 💪
What Came Out of It
- The transformation system - the foundation of positioning, rotation, and scaling objects - went through many iterations. While this is standard functionality for any engine, building it together with AI was unique in its own way. Modern models don't understand the concept of space. At that moment, the illusion of intelligence collapses: you realize it's just a token sequence generator - formally correct, but with no idea how it exists in the real world. In the end, I managed to make components responsible for visualization (whether static images or animated sprites) automatically sync with transformation changes, ensuring system responsiveness. 🖼
- Then came the turn for the physics system - again, no understanding of world concepts or how to connect it to transformations. After a day of "trial and error", I just sat down and wrote it myself. ⚡
- And what's an engine without its own editor? Me and AI created a set of tools to make the development process as comfortable as possible - here, AI was actually great, though I had to constantly nudge it to reuse existing code instead of rewriting all the styles from scratch every time:
- Object Inspector: Lets you deeply inspect any game object, view its components (like transform or sprite), and quickly tweak their properties. New components can be added "on the fly" or unnecessary ones removed.
- Asset Manager: A handy panel for managing graphics and other project resources. Assets can be easily dragged right onto the game scene.
- Sprite Sheet Editor: My special pride! A separate window where you can upload an image (sprite sheet) and visually "slice" it into frames for animation. You can create frames, change their size and order, and save this info. The sprite component then uses this data for smooth animation. No more tedious manual frame coordinate setup!
- Gizmo: An interactive tool on the scene for visually manipulating selected objects: move, rotate, scale. I paid special attention to making Gizmo intuitive and working correctly with the object's pivot point. You can probably guess - the model had no idea how to make this work. But after the 10th regeneration, it more or less worked. 🛠
So What's the Result? An Engine Born of Vibe and Code! 🎮
The result is not just a pile of code, but a full-fledged game engine!
Yeah, maybe it's not perfect, and its history has moments that make you smile, but it's mine, unique, built on pure enthusiasm, a flow of inspiration, and close collaboration with an AI assistant.
This engine is living proof that "vibecoding" with AI isn't just a buzzword, but a whole development philosophy.
It lets you not be afraid of experiments, boldly face mistakes (they're part of the journey!), and move forward with confidence.
It's a process where it's not scary to try new things, even if the intermediate results sometimes look... eccentric.
The main thing is the creative search itself, that unique "vibe" from creating something new, something yours, together with AI.
And this engine is the brightest proof of that. ✨
No-Drama Finale:
So that's the kind of vibe engine I ended up with.
Not perfect, not by the book, but mine - with bugs, memes, and a ton of pain to share with friends over pizza. 🍕
And yeah, not a single system is 100% finished: there are bugs and logic dead ends everywhere, and fixing them means it's easier to just rewrite everything from scratch.
Siccing an agent on refactoring isn't a great idea either: it'll run out of context or just lose the "thread of reasoning". 🤷
If you've read this far - respect!
Don't be afraid to experiment, argue with AI, sometimes wipe everything and start over.
That's the real joy of vibecoding.
And if you've had your own epic or funny moments with AI - be sure to share them in the comments.
Let's collect a gallery of vibey stories and laugh together! 😂
Top comments (0)