Admittedly, AI can now generate over 80% of the code for small and medium-sized projects. Building programs purely through natural language is no longer a fantasy. I firmly believe that within the next decade, natural language-driven programming will cease to be a pipe dream and become an irresistible mainstream trend.
Whether it’s Unreal Engine, Unity or Godot, game engines essentially offer a set of built-in rules and ecosystems for developers to build applications. It is entirely in line with the current trajectory of AI evolution for artificial intelligence to take over this implementation work. This marks a true AI revolution. Though it will demystify programming and strip it of its former aura, it is an irresistible general trend.
Programming is no mystery and carries no aura.
Neither of these things exist.
There is no quantifiable way of measuring the code required for “small and medium sized projects”.
The question “How much code is required for a small game?” has no real answer.
So you also pulled that “80%” out of the air.
Game dev has alway been part of AI development. One of the first things computer scientists did with AI was to make it program Pong.
My project, a flight simulator uses Godot, JSSBsim which are human programmed. AI programmed about 95% of the code that I would have had to program my self. It would have taken more then 10x longer for me to program that compared to the speed of AI agentic coding.
There are still limits to what AI can program but looking ahead it’ll cover a large degree of what usually is known as boilerplate code. So basically with the tools in existance (Godot, Unity etc.) combined with AI, any one can make their own games tailored to their own needs already today.
Of course not at a level of AAA games like GTA but it can get you quite far.
Oh but they do, for people who are completely clueless about programming. And now for some reason, they want to do programming without doing the programming thinking that will “demystify” it. It’s a cognitive dissonance of galactic proportions.
It’s fascinating how often the exactly 80% is uttered with a straight face by ai bros of all calibers, including your musks and altmans. How to take seriously someone who has never heard of, let alone experienced, the 80-20 rule.
I guess I could google it for you. But I don’t want to ruin the chance for you to discover it yourself
I can manly speek from the experience of my own project. I made a thread in this forum about it.
But it always comes down to the same question on how purist you thing that you have to be.
How deep into a computer can you dive befor losing your mind about it ?
Maybe this article can shed some light on what todays AI can do :
You said your own game was 95% coded by LLMs. Why would you have to Google your own game? Why wouldn’t you want to promote your own game when someone asks about it. That “ruin the chance for you to discover yourself” excuse is so cringe.
The developers describe how their game uses AI Generated Content like this:
*AI is used for entire code of the game, text, arts, sounds and music.
*
…
in just the first seven days of 2026, 230 games were released on Steam, and across that week 81 carried AI disclosures, ranging from upscaled screenshots to admissions that the entire game — art, voice acting, and writing — was built with AI
But that’s all the googling I’ll do for you today.
@coderbyte520 is clearly an AI Bro trolling. They literally made an account just to create that post, and are unlikely to come back to see wany replies, so engaging them and their made up statistics is pointless.
Have you been tracking the cost of your token usage? I’m curious how much your game will cost you to make using Agentic AI. I set up an Agent for my brother the other day to help him with his job search, and we hit the limits of free APIs just with the heartbeat pings during setup. He had to shell out money just to get it going.
Agreed. And this is why I think we are going to become artisanal programmers. Or glorified LLM debuggers.
For those of you who don’t know what @normalized is referencing here, it’s the lived experience that 80% of the code takes 20% of the time to write (LLM code), and the other 20% of the code takes 80% of your time to write.
So an LLM writing 80% of the code means it almost gets you there, but then what? Well, you put out slop like the game Codex Mortis, in which people say it’s fun for $4, but it’s not polished. Because, well they can’t polish it - that’s the 20%.
Some of the Reviews
It’s $4 bucks. You’ll get your moneys worth, but it’s super laggy, navigating the UI is abysmal with a gamepad and it gets really repetitious really quick. There is enough potential here to work with for sure, so if these issues don’t bother you and you want to disengage your thinking cap for a few sessions while drooling into a soft blanket by all means pick it up.
cheaper than a gallon of gas
My first impressions: 10% polish counts for 90% of the experience.
This game clearly lacks a polished, curated, concerted experience for the player. I wouldn’t rush to say it’s a terrible game of no merits. It works just fine, and I think in time the player would enjoy it just as well as any other low cost game on Steam.
Give it a chance, it’s not a horrendous unplayable experience, though it would certainly greatly benefit from more polish.
It made by Ai. I Couldn’t care less. i got my $4.00 worth of fun out of it. I am really looking forward to seeing how Ai enables people to create games that never could before just from whatever is floating around in their heads because they didn’t have time to learn to program, create models, etc. We are this far already. The next few years are going to be really interesting.
Fun addictive game. If you like immense of spellcasting and desctruction this is the game for you. As an AI engineer this is impressive done. Well done!
So yes, some people will buy a cheap, slop game. But those phone games have existed and been contributing to people’s increased toilet time for at least a decade. Making a toilet game with AI that a few dozen people played does not equate to success in my book.
I have no issue with being transparent about that. I have an account at Github.com with following subscriptions.
GitHub Developer Plan - Month: $4.00 USD
GitHub Copilot Pro+ - month: $39.00 USD
Plus I have one subscription at Anthropic for $20 + tax.
At first the Github subscription was an over kill as I never reached 50% of the monthly max. usage. Just recently they increased the ratio for so called premium requests. ( I didn’t see any notification that they rise the ratio but…). So I reached 97% last month. This month I’m currently at 82%.
But i’ve never payed extra, I rather wait for the month to end to continue.
Looking at what the best of their LLMs delivered so far I have no problem with what I payed them.
By the way I also donated three digits to Godo and will continue to do so, because I think it’s a very good project.
P.s. The reviews of the AI game matches that of many human made games too.
This is an issue with LLMs that is discussed less often. It creates a class delineation. $756/year plus tax isn’t a lot for a hobby for someone with a fair amount of disposable income. But the number of people with that kind of disposable income is going down daily. Which means it is a tool of the rich. Or at least becoming so.
I expected links to good 10x-ed games, not some second hand report of how much slop was “released” this year. We all know there’s plenty of slop everywhere. Quantity is one defining characteristics of slop.
Most people pay that for their Netflix, HBO, p***hub etc subscriptions. Which of, I have none.
I think that quality has its price. I could get some free LLM and run them locally but I’ve not seen any that are worth the electricity used. Also I have a clear goal, when reached I won’t need those subscriptions anymore. Hiring good programmer would be much more expensive.
Well, I do live in Switzerland. Here we’re all loaded.
But just like they can’t afford Gihub Copilot, they also can’t afford human programmer then. So what’s the difference ?
I don’t think we’re talking about “should I hire a programmer or use an LLM” but rather “I want to do game dev but I don’t want to have to learn programming” (and if you think that’s what these tools are capable of you’re massively overestimating them)
Then both would be an option.
Like I wrote above, I can talk from experience. I’ve used LLMs now for a bout a year to do my programming tasks. I know where their limits are and I see that the boundrys are pushed on a regular basis.
