AI-assisted code policy: "Literature Review"

[jasongrout]

I notice some conversation happening in various places around policies for AI contribution in Jupyter. I realize various subprojects are at different stages of this conversation, but I wanted to collect some links to conversation I see happening in various places. Think of this as a literature review.

• Apache Softare Foundation guidance
• Apache DataFusion policy
• Apache Arrow policy (conversation)
• Zulip policy
• Python policy
• Oxide policy (Bryan Cantrill)
• Qemu (some conversation)
• LLVM (conversation)
• ghostty
• gentoo
• Fedora
• Scientific Python discourse discussion
• Discussion from Scientific Python Developer Summit, 2025
• Somewhat related: Journal of Open Source Software (JOSS) comments on submissions in an era of AI
• Matplotlib
• Thoughtful blog post by Stéfan van der Walt (Scientific Python)
• NumFOCUS chat about AI policies, including a link to current policies for many NumFOCUS projects

Jupyter community conversation:

• 28 Oct 2025 JupyterHub Zulip conversation
• 18 Nov 2025 JupyterHub Zulip thread
• Oct 2025: Adding AI rules to the JupyterLab extension template: Issue, PR
• Jan 2026: Adding Agents.md to JupyterLab: Issue, PR
• Feb 2026: Zulip conversation around AI and the matplotlib fiasco

[manics]

Somewhat related, Research Software Engineering position statement on AI https://github.com/Academic-Data-Science-Alliance/rse-ai-position-statement?tab=readme-ov-file#readme

• Statement: https://github.com/Academic-Data-Science-Alliance/rse-ai-position-statement/blob/main/RSE-AI-Final_Statement.md

[choldgraf]

Here's one from Jellyfin (ht @stefanv who ht @yuvipanda)

https://jellyfin.org/docs/general/contributing/llm-policies/

[krassowski]

Discussion on GitHub Community page: Exploring Solutions to Tackle Low-Quality Contributions on GitHub

Also, a new report with some more context: https://www.theregister.com/2026/02/03/github_kill_switch_pull_requests_ai

[choldgraf]

Not a policy but this is a really interesting and infuriating action with a clawbot LLM in matplotlib:

• GH issue: [PERF] Replace np.column_stack with np.vstack().T matplotlib/matplotlib#31132
• Bot's angry blog post: https://crabby-rathbun.github.io/mjrathbun-website/blog/posts/2026-02-11-gatekeeping-in-open-source-the-scott-shambaugh-story.html
• Maintainer's own blog post: https://theshamblog.com/an-ai-agent-published-a-hit-piece-on-me/

[fperez]

Wow - thanks for flagging that @choldgraf. Wild interaction.

My starting point is that we must prioritize the well-being of humans in our teams, period. The fact that several maintainers had to spend that much time and effort treating a bot as if it was a first time human contributor we want to kindly welcome and onboard, rubs me very wrong...

I absolutely think AI assistance will play an increasing and even critical role in OSS communities, and I think it's possible to find ways for it to be useful.

But giving a bot access to our limited human bandwidth without bound is a recipe for disaster: the humans who wrote those measured and polite replies probably spent a LOT of time thinking them through and composing them, just for the bot to instantly turn around and generate more comebacks (and by going public/viral, using the amplifying power of the internet to push even more emotional work on the maintainers).

This is purely asymmetric information warfare with the bots, and humans have zero chance of winning that. We need policies that treat tools as tools, and humans as humans: with limited time and bandwidth, with real feelings and with wear and tear from these kinds of interactions that minimally bothers them, and often will just drive them away from a project.

Anyway, I need to noodle more on this, but thanks for sharing that, I'd missed it.

[manics]

A collection of policies, linked on the Scientific Python Discord:
https://github.com/melissawm/open-source-ai-contribution-policies

[jasongrout]

On the numfocus zulip, I also saw an article that has lots of links to what various people are doing.

[jasongrout]

I think this is a good case study of using AI to help open source projects. The author seems very conscientious, working well with maintainers to find and fix bugs with high quality reports. The focus area is fixing bugs in Python extensions.