My Experience Making Games from Scratch with AI
I have always been interested in making games, but I have never had a background in programming. I am almost 40 and work an ordinary factory job. About ten years ago I tried to learn coding on my own, but the amount of knowledge I thought I needed before I could build anything real felt overwhelming. Years later, AI changed that equation.
Читать эту статью на русском →
How it started
For my first experiment I chose ChatGPT. I started with the free version and used it to build a demo of a small children's game. I never released it because during development I realized I no longer liked the idea itself. But something important had already happened: I had proved to myself that I could create a working game.
I then started another project using ChatGPT Plus. Today I have a PC, a MacBook and several published games. Everything started on an old computer with almost no programming knowledge.
Why I chose Godot
I still had to learn basic concepts: how code is structured, how parts of a project interact and, most importantly, how to configure a game engine. I chose Godot because it actually ran well on my old PC. Unity and some other engines created installation or configuration problems, while Godot let me start building.
I also wanted Android and iPhone support. At the beginning I did not yet realize that proper iOS development and publishing would eventually require Apple hardware. I was simply experimenting and learning.
1. The idea should be yours
This may be my most important lesson. When I ask AI to invent an entire game, the ideas often feel too generic for me to spend months developing them. AI works much better for me as a tool for turning my own ideas into something real. That is why I started this journey: I had game ideas, but not the programming knowledge to build them.
2. First understand what you want to make
Once I have an idea, I create a project in ChatGPT and explain what I want the game to do. My first description is usually messy. Then ChatGPT asks questions, I clarify details, we discuss mechanics and eventually arrive at a clearer structure. I create an empty Godot project, package it and work from there.
The first working version is only the beginning. Adding features, mechanics, content, UI, ads or purchases, localization and platform builds becomes a long iterative process. For me, debugging is the hardest part. Debugging code you only partially understand is a special kind of pain, so even with AI a basic understanding of what is happening is extremely useful.
How I organize long ChatGPT projects
At first I kept everything in one conversation. After several days the context became huge and responses could take 10–30 minutes. Later I changed the workflow. When a conversation becomes too large, I ask ChatGPT to summarize the work and create a detailed handover note: what is implemented, decisions already made, important files and unresolved problems. Then I start a fresh conversation in the same project, upload the latest game and the summary, and continue.
I also tried separate chats for UI, advertising and gameplay logic. That did not work well for me because the conversations did not know what the others had changed. Sequential work in one chat, followed by a clean handover to a new one, has been more reliable.
AI makes mistakes too
Sometimes ChatGPT works extremely well for weeks. Other times it misunderstands simple instructions or claims it completed something that it did not. I have even received download links to files that were never actually created. These failures have not been common, but the lesson is simple: always verify AI output. Sometimes starting a fresh conversation is more productive than trying to recover one that has become confused.
Keep a notebook for ideas
New mechanics, modes and ideas appear constantly during development. Trying to implement every thought immediately makes a project grow out of control. Writing ideas down and returning to them later works much better for me.
Mechanics first, design later
I try to test mechanics, new features, screen transitions, saving and unusual cases before spending too much time on visual polish or monetization. If the game itself does not work properly, a beautiful interface will not save it.
Create developer accounts early
If you genuinely plan to publish, do not wait until the game is finished before learning Google Play or the App Store. Developer accounts, testing tracks, reviews, store requirements and technical configuration take time. Starting early means some publishing work is already behind you when the game is ready.
AI can help with much more than programming
I have used AI for game mechanics, programming, debugging, UI work, writing, localization, visual design, image preparation, Google Play and App Store materials, testing, analytics interpretation and release preparation.
Can you make a game without knowing how to program?
My answer is yes. But my experience is not typing “make me a game” and receiving a finished product a few minutes later. AI dramatically lowered the barrier to entry, but I still have to learn, make decisions, test results and fix mistakes.
For me, the biggest change is this: ten years ago there was a huge wall of knowledge between having a game idea and being able to build it. Today there is an assistant between those two points that I can ask for help exactly when a problem appears. For me, that difference was enough to make game development possible.
The games that came from this process
Cipherix and HorP are two published Dexiges projects built with Godot through this AI-assisted workflow.