Unfortunately the use of generative AI tools is no longer just a personal choice when they are used in open source projects where many people are all collaborating. Which is why we need a discussion and official policy as a community.
Many maintainers do not want to read AI-generated PR descriptions or comments; they are often needlessly verbose, they can contain false information/logs that "look correct" and often times it takes a lot of back and forth to get answers to questions because AI models struggle to understand what is being communicated in some cases. I had an interaction not long with an entirely AI-opened issue and comment replies: https://github.com/apache/nuttx/issues/19135. This is a waste of my time and it is incredibly frustrating to read. It should not be allowed at all. It is basic respect to communicate yourself with your fellow humans if they are giving you their time for a review. Of course, machine translations don't fall under this umbrella which is why they are excepted in the policy. Then there is the code itself, which can also contain hallucinations which "look correct". That makes it harder to review. AI (just as humans do) can create non-idiomatic or very verbose code as well. This is also harder to review. There are also the copyright issues where AI produces code that isn't compatible with our licenses. Then there are also cases where contributors fully generate AI PRs and admit they were not able to do the task without the AI and were heavily reliant on its output for their patch. This creates a ton of code that is entirely unmaintainable, because the contributor doesn't understand the patch they're submitting. I know that isn't the common case, and we have many contributors who are using AI responsibly, but we are still getting patches like this and we need a policy about it. I believe the Godot policy addresses all of these concerns and is not particularly restrictive towards those contributors who have already been using AI responsibly (for large refactors, as a guiding assistant, etc.). It gives us moderation ability for these extreme cases, it does not forbid the use of AI as a blanket ban. Matteo On Fri, Jul 3, 2026 at 1:35 PM <[email protected]> wrote: > I agree on what Xiao said. IMHO the human is always reposible for the > code, regardless of the tools used. Adding tags about the code creation > tools doesn't seem to add any valuable info. The code needs to be good > quality and the responsibility is on the author anyhow. > > Having said that, I don't strongly oppose adding thise tags, it just looks > useless to me. > > Xiang Xiao kirjoitti perjantai 3. heinäkuuta 2026: > > Actually, I don't care the code is created by humans or LLM, all patches > > are equal to me and must pass my quality bar before I approve it. > > > > On Fri, Jul 3, 2026 at 2:44 PM Xiang Xiao <[email protected]> > wrote: > > > > > Whether or how a developer uses AI is a personal decision. What I am > > > against is creating a rule to forbid the patch the developer generated > with > > > AI. > > > > > > On Fri, Jul 3, 2026 at 2:33 PM Michał Łyszczek < > [email protected]> > > > wrote: > > > > > >> On 2026-07-02 18:54:44, Sebastien Lorquet wrote: > > >> > Everything about "ai" is wrong and problematic, even if you want to > > >> tolerate > > >> > one aspect, it stays unacceptaoone has asked me either, but I 100% > > >> agree with those. > > >> > > > >> > -quality is unacceptable, > > >> > > > >> > -environnemental impact is unacceptable > > >> > > > >> > -ethics are unacceptable > > >> > > > >> > -it is completely unsustainable on the financial and industrial > levels > > >> > > > >> > -above all, it renders humans incapable and addicted to a > centralized > > >> system > > >> > owned by billionaires > > >> > > > >> > -it destroys communities and cultures > > >> > > >> Noone has asked me, but I agree with Sebastian here and I am rather > > >> against using LLMs myself too. > > >> > > >> I don't like that my code is used by corporations to create their > product > > >> any > > >> money that they use to addict you to the tech. It's one thing when a > > >> human reads > > >> my GPL code and uses very similar code to implement something in the > > >> company. He > > >> did benefit from it somehow, he got smarter, maybe he saved his job, > > >> whatever, > > >> but he's the primary benefactor. > > >> > > >> With LLM? Company benefits first. And the dude may even loose his job > > >> because of > > >> LLM. And I want for people first to benefit from my work - not > > >> corporations. > > >> > > >> Also LLM makes you dumber. Especially if you are learning. I use LLM > > >> sometimes > > >> as search engine or example generator. But even with that, I can feel > my > > >> brain > > >> getting lazy and poking me to use LLM for that quick solution. From > one > > >> point > > >> it's cool to get solution faster, but on the other hand, we will loose > > >> ability > > >> to solve problems on our own. > > >> > > >> I can see benefits of LLM. Like the time you guys used it to refactor > > >> stm32 > > >> source file to be more structured. Yeah, I think that's good use for > LLM. > > >> But > > >> there are just much more bad use cases. > > >> > > >> Can't speak for anyone, but I avoid LLMs mainly for my and my brain's > > >> sake. I > > >> believe LLM will cause more harm than good. Especially in the future, > when > > >> experienced programmers pre-llm era will start to die out. > > >> > > > > >
