Rémi Verschelde (Rémi Verschelde), one of the key maintainers of the open source game engine Godot and co-founder of W4 Games, said that the growing flow of low-quality Pull requests generated by large language models are exhausting and demoralizing for Godot developers reviewing changes. According to in Remy’s words, the Godot project prides itself on being open to new contributors and giving any user the opportunity to influence the development of the engine. Godot maintainers spend a lot of time helping to get pull requests up to par, but with the current barrage of questionable changes, Remy doesn’t know how much more the team can handle.
There are currently over 4,600 open pull requests in the Godot GitHub repository . Adriaan de Jongh, director of game development studio Hidden Folks, characterized the state of AI-generated garbage pull requests for Godot as a complete mess and a huge waste of time for reviewers who are forced to waste time parsing changes that often do not make sense, are provided unnecessarily verbose descriptions and are sent by developers who do not understand their own patches.
With the advent of AI tools, maintainers are forced to additionally determine whether the code is human-written or automatically generated, as well as to understand whether the author understands the code he sent. You also have to figure out whether the authors tested their code or whether the presented test results were just fabricated by AI. In addition, Remy does not understand what to do in a situation where, when asked about the use of AI, patch authors answer “I only used it for description, because I write bad English,” which does not allow unambiguous classification of contributions.
It is noted that maintainers have to double-check pull requests from new participants several times a day, doubting their quality. Remy believes that the only solution to maintaining the previous approach to development is to increase funding for the project in order to pay additional maintainers who will parse the AI slope.
The Blender project is experiencing similar difficulties, which in early February began development of a policy regarding patches created using AI. Previously, rules for such cases were adopted by Linux Foundation, Fedora, GNOME, Firefox, Ghostty, Servo and LLVM. In January, GitHub began discussing the implementation of measures to limit low-quality pull requests generated in AI assistants, submitted without manual review, do not meet quality requirements and create serious difficulties for maintainers.