Factorio Forums

How is the mod scene going with AI these days? I have a dream for a super sleek and fast elevated Maglev rail system. My internet is unfortunately too slow to get anywhere close to doing it myself...

See some inspiration here: Couldn't get this one to run: (AI made mod)

there are various point of view , sometimes very opposite about AIs in general, and that is also present in Factorio. I'm not a person who is going to hate on AI , i try to have a rationnal approach, still i think for Factorio modding in particular, there isn't a real good AI i know that can do "everything" as you seem to wish. I tried using AIs to make mods, but i found it better to use them to understand how to make a mod , rather than to make the mod themselves, they don't have good reference on what is availble in factorio API, they make mistake between 1.0 1.1 2.0 space age and overall they are pretty unreliable for task like "can i do this ? or how do i do this ? or can you do this for me ", rather, you can ask "what does this do ?" and giving snippets of code you don't understand ( respecting the license ) , you will have an explanation, which is very often correct, it allows to spot mistakes in what you did, you can also ask utilities stuff, like i have a list and i want to remove only the elements that are also on this other list unless they are present on this 3rd list and i need to do that on those 4 files everywhere needed, AI speed-up stuff with less risk of mistake if you don't rely on them having a perfect, nor even good "game internals knowledge" and follow their own hypothesis.
You can try to "reason" depending on the quality of the AI model you use, try to have a discussion where you explain the AI your goal, or intermediate goal, or tiny step, ask it how it is done by people who know what they are doing, and why ,then it's easier to further ask precisely what you want the AI to do, instead of "the whole mod", you ask for smaller parts that you know you need because of the previous dicussion, where you try to spot if there is inconsistency, like the AI telling you A is the better method then giving intruction for how to make B; and you test if those intermediate step produce the expected result, that the AI doesn't fallback on a proposition C that look like it does the job but not really. If you don't ask it to explain to you what the code is actually doing it's impossible i think for such project in 2026.

There are more expensive AI agents than those i tried, but when i try to document on how it works, i am not convinced there is yet a stage where you can ask for "such a complex mod" to be done reliaby "at the end" without much more intermediate validations. An AI agent as far as i understand very much risk going to otherwise bash his money longer on problems like version inconsistency, never understanding the root cause and developping many alternative wrong theory and test them over and over. Unlike some other development , a factorio mod often time needs to be reloaded and tested humanly to know if the incremental changes and addition by the AI are working.

I feel your "best bet" is to take a month or so for it minimum, and validate a plan for each week, like 1rst one you load the mod and the technology and the icons and the tech, but it's just a placeholder non animated graphic, the week after because it takes time to do the back and forth with the AI, adding a thing, correcting the mistakes, learning in the way so you don't need to always reload the game to spot them in the AI code, you still learn the langage, the data structure the API and so on. I think for the graphics themselves you need no less than a week, and it probably will still look like placeholder, but undertanding which pieces are needed, getting the format right the dimension the proper set you need for a full animated trains, and making mock up version is already going to take time and require intermediate human verification, you will most likely need to correct the AI based on what you visually observe is going wrong, that's more the wiring parts, you need also at least a week for this, the animation in game and the tweaks and corrections for the graphics to be actually displayed as a functionnal train and not just there somewhere.

And you also need a week at least for the polishing, making sure it doesn't break with the other mod you like to use, make sure the performance arent terrible because of leftover scaffold or debug code, typos, stuff you left as "to do" making the mod page on the portal, and for such a mod , i expect also dealing with the early feedback for such a mod, people always find edge cases, some you may want to solve.

Obviously i'm not an experienced modder, i tried to touch things here and there, and if AI were better at helping or i had developped a better method of using them i could have touched some more, i felt the limitations, i'm not doing a generic "AI bad argument" here, i'm saying it's actually much more work than may be anticipated from the fallacious promises of companies selling chatbots ^^ When i say "1 week per stage" it's ambitious and approximation, it's more like "at least a week" or so in my mind, and obviously depending on how much time you can spend and if you are a programmer already, or if you have a background that allows you to learn and remember fast the new concepts. Also Discord can be quite useful, like if you really can't find the answer to a specific stuff even with AI, go ask real humans, but you need to narrow down the problem better than "i asked chagpt once and it didn't immediatly work" , otherwise you're asking someone to review the full mode and finish what could be 95% of the work (and real humans rarely do that x))