🤖 Dealing with LLMs / "AI" slop / maintenance efficiency / burnout prevention

[webknjaz]

👋 Many of you must've noticed already how today's spam networks led by sloperators can automatically waste meaningfully invested maintenance time on scale. People who know what they are doing can often use modern LLM/agent-based tooling to help them and the projects achieve new heights efficiently. These are objectively great at being our rubber 🦆 or a rabbit hole exploration partner.

🙈 However, there's a category of contributors who when exposed to such tools without prior experience and education, are known to hurt projects and their maintainers, albeit not on purpose. I've actually started tracking examples of such behaviors for myself to observe patterns: 📅 Procrastinating in public 😵‍💫 (view).

And so I've been looking into how other projects deal with these, collecting examples, researching points of view etc. Here's a few resources that attempt to collect and/or visualize different approaches across FOSS:

• https://github.com/chaoss/wg-ai-alignment/tree/main/moderation#readme
• https://github.com/melissawm/open-source-ai-contribution-policies
• https://redmonk.com/kholterhoff/2026/02/26/generative-ai-policy-landscape-in-open-source/
• https://github.com/orgs/community/discussions/185387

With that background I've initiated my own attempt to come up with a set of guidelines for one project (pip-tools was chosen as a playground), having in mind that I'd like it to be scalable and could copy into many other places that so desperately need something like this. I asked my friend @sirosen to lead that effort and invited more people of varying backgrounds to contribute their thoughts. Here's the discussions with ideas/contexts/justifications:

• [policy] Good-faith agentic contributions and LLM use, and avoiding death by a thousand AI slops jazzband/pip-tools#2278
• Introduce an initial LLM Usage Policy jazzband/pip-tools#2318 (this one ended up being merged on Feb 28)

I'd like to encourage that @aio-libs and individual projects within to take this prior experience and adopt it as a base. It can evolve but I think it's a good place to start. Here's what landed in the contrib guide (the one for humans) and we can copy in some shape:

• 📚 https://pip-tools.rtfd.io/en/latest/contributing/#llm-generated-contributions

As for the future improvement ideas, thei include:

• ✌️ automating trust scoring with something like https://github.com/marketplace/actions/good-egg-trust-scoring-prs — it can be used as a nice early indicator for the maintainers to be able to recognize potential slop and respond with links to the docs, for example
• some projects reported great success with auto-closing PRs that don't follow the template (which is what things like open claw do due to the way they use the API for PR submission)
• investing into context files (agents.md/claude.md etc) to make it easier for respective tool users to pick up the (often implicit) expectations / tribal knowledge

cc @achimnol @bdraco @Dreamsorcerer @jakob-keller @jettify @Nothing4You @mjpieters @aio-libs/admins @aio-libs/sprinters @aio-libs/triagers