On October 7, the official QGIS repository formally merged a proposal numbered QEP428, which does not permit any code generated with the assistance of LLMs, AI, or agents. Actually, this is not the first time QGIS has done this. I have roughly looked through the discussions and mailing list, and today I will talk with you about this matter.

Details of the Rules

There are two formal clauses.

First, any code contribution assisted by LLMs, AI, or agents is prohibited from entering the QGIS repository; second, if a violation is discovered afterward, the code may be rolled back, and sanctions may be imposed on the contributor's permissions.

The proposal was put forward by core developer Nyall Dawson, and the corresponding pull request is QGIS-Enhancement-Proposals#387. After the discussion period, it went to a vote and was finally merged on October 7. At present, there are both voices in support and opposition.

However, QGIS officials do not intend to keep hand-written code forever. Rather, this is a choice they have had to make under QGIS's current development model. AI uses such as autocomplete and translation will be opened for separate discussion later.

Previous Policy

Earlier, QGIS actually already had an AI policy, numbered QEP408. The idea was that AI could be used, but the output must be manually reviewed and publicly declared with labels such as Assisted-by. The idea sounded good on paper, but for a large project like QGIS, this process did not really get off the ground in practice. According to community statistics, about 50 PRs were labeled AI Assisted, of which about 40 were ultimately closed without being merged.

According to QGIS maintainer Régis Haubourg, after external contributors submitted a large number of PRs using AI, the review volume far exceeded the upper limit of manual review. As a result, the review queue kept growing longer, and code quality could not be kept consistent. This is something the QGIS community does not want to see. Traditional open-source software relies on collaborative networks among people, with everyone collaborating through the community. But code produced by AI pipelines continues to take up reviewers' time, and the space left for human contributors is squeezed out. More troublesome is that people who submit patches with AI often do not learn the project itself, and find it hard to grow into the next generation of reviewers, so the community's relay of skills will slowly break down.

PS: I think this is put somewhat euphemistically. The core can be summarized as: Why should I read line by line and test one by one the garbage you generated with AI?

How Is It Enforced?

Some people in the discussion worried that the policy is written too broadly. Does smart completion in an IDE count? Does incidentally seeing an AI summary while using a search engine count? Does chatting with ChatGPT about a design and then writing the code by hand count?

Even Rouault's response was: the community knows it will overcorrect, but this is a transitional policy. Two years ago, before this AI wave, QGIS was not worse, so it is better to tighten first.

Others followed up asking how it will be enforced. The current response is: rely on trust and the deterrent effect of the policy; after the fact, code will be rolled back and permission restrictions will be imposed on developers who do not comply.

Summary

QGIS is one of the more influential projects in the open-source software community. The discussion of its policy has already gone far beyond the GIS circle, and the final outcome may affect how all open-source software is developed in the AI era.

The current contradiction is that AI has pushed the cost of submitting a PR very low, but has hardly reduced the cost of serious code review and maintenance. The core maintenance capacity sustained by volunteers is being crushed. Is QGIS's decision good or bad? Honestly, I cannot answer. You are welcome to share your views in the comments.

References

  1. Add QEP428: No ai contributions permitted: https://github.com/qgis/QGIS-Enhancement-Proposals/pull/387
  2. QGIS AI tool use policy (current text): https://github.com/qgis/QGIS-Enhancement-Proposals/blob/master/qep-408-ai-tool-policy.md
  3. qgis-developer mailing list follow-up discussion: https://lists.osgeo.org/pipermail/qgis-developer/2026-September/068576.html

Tags: QGIS AI-assisted coding Open source governance