Rémi Verschelde, maintainer of the open-source Godot game engine project, has recently taken to Bluesky to express his frustrations at the added burden placed on open-source project maintainers and developers by the increasingly common practice of submitting PRs with completely or at least mostly AI-generated code. The gist of his issue with this code is that it often contains errors and oversights that result in maintainers wasting time not only reviewing sloppy code but also discussing the code with the developers responsible for the contributions. In the post, the developer notes that in many cases where code is obviously written entirely by AI, the proposed changes "make no sense," and goes on to question whether the code was tested, if the contributor made up test results, or even whether they understood the changes they were proposing.

In addition to questionable code, the explanations for the contributions are also excessively verbose and difficult to understand, further adding to the workload for maintainers. This isn't the first time we've heard directly from developers that AI tools have been more of a hindrance to productivity than anything else. Shortly after EA announced its AI pivot, workers at the gaming giant claimed that the AI tools that had been pushed by management for a year at that point were costing developers time instead of increasing productivity. There have been similar discussions in the Blender development forum, where one user proposed implementing an AI contributions policy that would only allow AI-generated code in a limited capacity, only when disclosed, in cases where the developer takes full accountability for their submission, and when they understand the utility and functionality of the code.