A recent pull request in the QGIS‑Enhancement‑Proposals repository introduces QEP428, which would prohibit any AI‑generated contributions to the QGIS project. The discussion, captured in the pull request thread, shows a range of opinions from strong support to concerns about the policy’s breadth and potential unintended effects.
Supporters argue that allowing AI‑assisted code erodes the trust contract between reviewers and contributors. One participant noted that when they used AI to fix bugs, they “didn’t realize the harm this could cause to QGIS’s already fragile reviewer community” and that the practice breaks the expectation that contributors know what they are doing. Another comment warned that reliance on AI prevents contributors from building skills, saying they “can never really hope to become reviewers themselves, and the community stagnates.”
The proposal is framed as a corrective measure to an earlier, less restrictive policy that failed in practice. Contributors pointed out that the previous “extractive tag” was never created, about 50 pull requests were labeled AI‑Assisted, and roughly 40 of those were closed without being merged, with several others closed but never tagged. Because of that perceived failure, the authors of QEP428 advocate starting with an “ultra‑restrictive” stance—a closed door—that could be opened later on a case‑by‑case basis after further discussion.
However, many commenters expressed worry that the policy’s language is too vague. Questions were raised about whether the ban would apply to issue comments, pull‑request descriptions, or the QEP repository itself. Others asked whether tools such as GitHub Copilot, Claude Code, or LLM‑backed autocomplete would be forbidden, and whether non‑LLM AI features like JetBrains’ model‑ranked autocompletion would fall under the prohibition. One participant warned that the broad wording might “throw out the baby with the bathwater,” potentially blocking useful AI‑assisted activities such as using a language model to gain deeper code understanding, to simulate reviewer feedback, or to brainstorm designs.
To address these concerns, some suggested revising the policy to replace the generic “AI” term with a clearer focus on “written” contributions, explicitly allowing activities like searching for error messages, brainstorming, or using translation tools, while providing concrete examples of what is and is not permitted. The measure of compliance was described as asking contributors whether they, in good conscience, wrote the code themselves.
As of the latest update in the thread, the proposal has moved past the discussion phase and entered the voting stage, with participants urging that the vote remain open through the week to allow broader community input.