The QGIS project has opened a discussion on whether to bar contributors from using AI‑assisted tools when preparing code changes, documentation, or any other project‑related content. The proposal, labeled QEP 428 and submitted as pull request #387 in the QGIS‑Enhancement‑Proposals repository, argues that the current “human‑in‑the‑loop” policy has not prevented problems in the reviewer community.
According to the discussion thread, one contributor described submitting two AI‑assisted pull requests that fixed bugs but realized afterward that the approach could erode trust between reviewers and contributors. The author wrote that relying on large language models creates an illusion of competence, prevents skill development, and may discourage contributors from becoming reviewers themselves, leading to community stagnation.
Other participants noted that the project’s security advisories are already produced using AI tools, though most relate to the project’s websites rather than the core QGIS codebase. Some commenters expressed concern that a broad ban could also rule out beneficial uses such as language‑model‑powered IDE autocomplete, design brainstorming, or using LLMs as a simulated reviewer to catch obvious mistakes before human review.
The proposal’s authors intend to start with a fully restrictive stance, then later consider opening exceptions on a case‑by‑case basis after further discussion at the upcoming user conference. They characterize QEP 428 as a correction to an earlier policy that, in practice, was not implemented and resulted in the loss of at least one high‑value contributor.
As of the latest comments, the proposal has moved past the discussion period and entered a voting stage. Supporters argue that a clear prohibition is necessary to protect the reviewer community, while critics ask for more precise language to distinguish between disallowed AI‑generated contributions and permissible AI‑assisted activities like translation or error‑message searches.
If adopted, QEP 428 would require contributors to certify that any code, comments, or issue content they submit was written without the assistance of large language models, marking a notable shift in how the QGIS community approaches emerging AI tools in open‑source development.