Seemed cool as hell! :D
Until I saw you used AI for stuff. Don't care if you edit what it slops out afterwards. I dropped Stellaris after over 1,000 played, and had it removed from my account over them adding AI content.
AI used = Never touching it.
Copilot and Claude *are* generative AI. And I agree with OP. Why even bother doing so much hand-made work only to use bullshit AI to do translation (making the non-english game experience abysmal), debug the game (adding more bugs), and *write items and enemy effects for you*? (what even in the gameplay mechanics is even yours at that point?)
I used AI for modeling and debug the most complex subsystems: for example i have no idea how a immune system of a bird works in detail but the AI does i can manage this on an high level by connecting the outputs and inputs of the various subsystems, i come from a 10+ year coding experience so i know what i'm doing ahaha is not just mindless vibe coding.
It's purely a time based problem: making all the complex and intricated economic, physics, chemical and biological simulations that use real world science will take me as a solo developer years of study and several university degrees, time i would rater spend making this game :D So on a high level scale i just need to know IF a subsystem works and provide the correct input and outputs to the rest of the simulation, the implementation details are not that important.
Game translation will be done manually in both english and italian and it uses a lot of slang, technobabble and neologisms that the AI just can't do, i don't plan to use it to translate it to other languages since i don't speak them and i can't verify if the translation is correct, but i leave them to any native speaker who want to translate it!
All the placeholder dataset of items and spells are being slowly implemented and rebalanced manually one by one, at the end of the two year developement cycle basically no AI generated data structure will go untouched.
Also no AI graphical or musical asset is or will be used, AI usage is limited just for code and DB and all the code will be made public on release if anybody want to reuse the subsystems.
Research is not hard. I wanted to make a small fantasy setting for a project I am doing for fun, and I wanted to have the elves in my setting be based on Otto-Turk cultures and mythology.
Instead of using a machine that only knows how to string sentences together in a way that imitates the most commonly used sentence structures. (which is why they are called LLM's, Language Learning Models.) I just spoke to two of mt friends who were born and raised in Turkey a couple questions then used Wikipedia searches for the rest.
It took two afternoons, where I did an 8 hours shift in the morning, to make a faction of elves who have a culture and feel that is VERY Turkish/Mongolian,but wasn't a copy and paste job. If you cannot just look up how a bird's immune system works. That is on you.
It's not just look up just how a bird's immune system works: the game needs to model every imaginable system present on Earth, from tire tension of monster trucks to the effect of moon rays in the navigation of moths, based on moon position in the sky and lunar phases. The AI is making every possible submodel with a common schema defined so that it can interact with all compatible submodels by sharing data and state infos.
My job is to organize all this obscene mass of metadata into a coherent procedural world, and this can be done just by hand and by a human since the context how the whole code exceed what AI can store in memory withouth crapping herself: the clanker's job is to make in a hour a model that will take weeks of study and manual coding, and every bits of code written by AI is double checked and corrected by hand.
Just another example: in the multiverse view, when zooming out at the maximum level, i've added for some eye-candy some four-dimensional hypercubes mapped in 2D, i have absolutely no idea or time how to learn how to model four dimensional objects in javascript and i will never do in two lifetimes, but the AI wrote a method do draw a procedural xray tesseract in like 30 seconds.