I'm just a scrub when it comes to coding, but using AI i can bring a game to maaaaybe 50-60% completion
Would you, as a real developer, take a commission to polish the game to completion?
Or would it be better just to use the AI game as rough demonstration of what I want and rebuild the game from scratch?
just simple HTML5 Web Browser Games,
I'm going to be looking to hire in 2-3 months and wondering whats smartest way to do it
As a new dev that built a new game 100% from scratch with AI. There are some things to keep in mind but for the most part it worked out well. The magic is in the prompt. You have to be patient, and carefully explain your vision to the agent as well as details on every little thing - the more detail the better. In my example, I made a new RTS game with all new mechanics (so much so that it wasn't compatible with existing engines), so I had to prompt it little by little over the course of many weeks (about 5-6 weeks total) with hundreds of milestones (sessions where one thing or related things were implemented one at a time).
For my background, I'm 100% an amateur game dev but I have a very strong background in computational physics and applied math so I knew exactly how I wanted the code to work; I could have coded it myself but it would have taken 50x the time! The end result of my game was around 170k lines of code in the src folder of my repo (my repo is private, sorry).
As for those that consider it unethical, I don't think it is as I use human-made assets bought here on itch (music and sounds). The art is currently all simple mono-chrome vector graphics but if I want something better I'll want to buy asset packs or hire people.
And as for "fixing" the AI mess, it's not so much as fixing but rather being very clear how things should be managed and even asking the AI for suggestions on how to manage the project! The magic is in the prompt, bad prompt = bad game that's very hard to fix.
a real developer, take a commission to polish the game to completion
What do you think is the hour rate of a real developer? And how long do you think it would take to "polish" the game? Or rebuilt your prototype, for that matter.
You might find an amateur hobby dev that is willing to take on this task, but you will not find a real developer that works for pocket money.
You use ai. Ask an ai. They are good for this type of information gathering and for getting estimates. You should still doublecheck any crucial information, because ai can hallucinate and just make stuff up, but reading information random strangers claim on the internet is not exactly reliable either.
This was the answer I got:
cost estimate on building a small computer game, if you hire a developer for the task.
A small computer game typically costs $10,000–$100,000+ to build with a hired developer, depending mainly on scope, art quality, and whether you hire a freelancer or a studio.
--
Using AI to replace employees, including developers, is looking more and more like a scam by ai companies. They try to sell their services and that service is costly and can not deliver, only imitate. If imitate is all you need, you might be fine. Or if there are no great risks. A hobby developer creating a game does not have to deal with customers running away, because of all the bugs. Some will not visit because of the ai, but it's a hobby, who cares about success, as long as you have fun making the game you could not have made without ai. I see this similar to all the sophisticated game engines. Boiled down, many a game is just an elaborate level for rpg maker or another engine.
I see this similar to all the sophisticated game engines. Boiled down, many a game is just an elaborate level for rpg maker or another engine.
I agree that there are scammy devs out there just to make a quick buck and poison the pool of games by making actual "effort-put-in" games buried in a sea of slop.
But some AI coding people like myself do try to make all new ideas come to life - with the right prompt! I did have to make my own engine (and I built it with AI) because the existing ones (I checked several) do not work with storing history state buffer to calculate earlier positions, hp values, entity states, etc. in a complex past-present visual renderer. So not all game applications involving AI bad!
There are probably a number of people like me who don't have the time do invest a lot of our time and effort into making our ideas come to life. In my case, it's because I have a full time salary job and game development is a new hobby for me.
I do not think you understood my points.
It is the ai companies, that show scam like behaviour. They promise their ai can do things, but it often can't. And companies notice this months later. That ties into the problem, that an ai made game that needs human polishing will be more costly to polish, than it would cost to make the game from scratch. If it needs polishing, that is. A smallish project might get away with being ai made. Depends how you used ai. As a gloryfied spellchecker and autocomplete, or to substitute all the knowledge needed for game development.
The risk for a hobby game project is minimal. Except for some people avoiding it, becaue of the ai. You are not a company that loses established customers because you switched to ai.
You did not even switch at all. You use ai as an enabler.
Just like people use game engines as an enabler.
So not all game applications involving AI bad!
That is a different discussion and that statement is trivially true. But if you were a team of three and you fired the other two, to replace them with ai, I dare say, that would be seen as bad by many people and for different reasons.
I would compare it to learning a new language. I would count strong fundamentals as knowing data types, loops and objects. I learned a lot from professors but the same info that is in classrooms is online. I think you have to be a bit obsessed with tech to grasp it quickly. Just start from the very beginning with a "hello world" program and move on from there.
Yes, a developer can take a paid commission to finish an AI-assisted HTML5 game. Whether repairing the current build or rebuilding is faster depends on the actual code and what still needs to work.
I can offer a small paid first milestone: inspect one playable build and its source, reproduce the main blockers, check your 10 MB ZIP and 2 MB per-file limits, and give you a prioritized repair-versus-rebuild plan with an acceptance checklist. We would agree the fee before that review; any coding would be a separately scoped, fixed-price milestone.
I use AI-assisted coding and verify changes by running the game and repeatable checks, so I want to be clear about that upfront. If you're still planning to hire around October/November, is there one game and unfinished mechanic you'd want assessed first? A short description or public demo is enough to scope the first step; source can be shared privately after we agree how to work together.
In my humble experience, AI tends to make code longer with each pass and tends to make over defensive and verbose code. I find AI written code unreadable. But I can also feed AI written code into AI and have it rebuild the codebase to be human readable. So it's kind of a wash. One problem with bigger projects is AI tends to iteratively make the same code longer and longer and longer and difficult to maintain. But you want to know the best way to manage that? Also AI.
Here is the deal. I think we are in a new era. You can let people complain or make up fake ethical dilemmas to virtue signal their avoidance of AI. But it isn't going anywhere and every game going forward will be using it to generate code whether they tell you or not.
But the real important thing is it's more fun to build with AI. You can do more and do it faster and that lets you be creative.
But the real important thing is it's more fun to build with AI. You can do more and do it faster and that lets you be creative.
That really depends on how you use it. AI(P) undeniably makes everything faster, but it doesn't necessarily make it more fun or more creative.
If you let AI do all the work for you, then there's no creativity on your part and it's more prone to its most common flaws like artifacts or hallucinations (despite its huge improvements), but if it's assisted for the difficult parts and you add your touch for the revision and where you do best, then there is some creativity.
Or would it be better just to use the AI game as rough demonstration of what I want and rebuild the game from scratch?
I'm going to be looking to hire in 2-3 months and wondering whats smartest way to do it
AI is changing so quickly, I expect within a year you will be able to get closer to 80 or 90% complete versus your current estimate of "50-60%"
So I would just keep "learning the tools," doing lots of experiments and rebuilding your same idea over and over again (sometimes regeneration with new learnings is better than iteration, especially with these new and developing tools).
But I do not think AI will ever generate a "100%" complete output. If it could,the bar for "what complete is" would raise. So as you are learning the tools, also ask the AI to make tools for you to manually improve the things it seems to struggle with that bother you (for instance art tools, animation tools, level design tools, etc)
It is trivial to generate bespoke tools for your own game which is the real true history of game development before the unitool engines took over.
Normie Woes with AI
What happens when a completely ordinary person tries to make a game with AI?
Normie Woes with AI is a short, chaotic indie experiment about using AI to turn random ideas into an actual game.
Expect questionable decisions, strange AI-generated solutions, unexpected bugs, and the eternal question:
“Is this actually a game yet?”
Made as an experiment in seeing how far AI can help an ordinary beginner go with game development.
Built with AI. Tested by a human. Somehow still works.
This project uses AI assistance during development. The extent of AI involvement may vary between code, ideas, text, and other development tasks.
ChatGPT was used as an AI development assistant.
