DEV Community

Cover image for Open Source and the agentic wave
Floor Drees
Floor Drees

Posted on

Open Source and the agentic wave

#opensource #ai #postgres #kubernetes

At FOSDEM PGDay last week, Jonathan Gonzalez and I presented a lightning talk about how the CloudNativePG project has been flooded with AI contributions lately, as a way of catharsis. Below is the script (somewhat).

Coincidentally this weekend the project adopted a policy regarding AI-generated content, inspired by the Ghostty AI Policy (the new terminal project, spearheaded by Mitchell Hashimoto, who previously founded HashiCorp) and the broader educational efforts of the CNCF and the Linux Foundation regarding the responsible use of Generative AI in open-source development. Summarizing the need for the policy:

While we recognize that AI-assisted tools (such as Copilot, ChatGPT, or Claude) can be powerful aids for development, they also facilitate "low-effort" or "random" contributions that increase the burden on maintainers without adding proportional value.

Reading out AI generated PRs for fun an profit

Open sourcing your project, and/or donating it to a foundation is usually good for the longevity and sustainability of the thing. Provided the problem you're solving is relevant and contributors know how to find your project, you're likely to benefit from more eyeballs and external perspectives.

But with great visibility comes... the agentic wave: AI generated PRs and bots in your community channels.

Quick note on the project we’re talking about specifically. CloudNativePG is a Kubernetes operator for PostgreSQL. The project was open sourced in April 2022 under the Apache 2.0 license and has collected almost 8000 stars on GitHub to date, which tells you something about the amount of attention the project attracts. In January 2025 the project joined the Cloud Native Computing Foundation as a Sandbox project. The CNCF has three maturity levels: sandbox, incubating and graduated. With each level usage and visibility increases.

In the spirit of the entertaining talk “Kubernetes Maintainers Read Mean Comments” - where the speakers do just as the talk title suggests: read the mean comments on GitHub issues and PRs - we thought we, who are at the forefront of CloudNativePG triage, could read out the worst AI drivel we encountered masquerading as contributions, for "fun and profit".

Let’s look at some gems.

Copilot PR

“I used copilot to implement this fix as I do not have any Go knowledge”
No comment.

One PR changing 100 files and adding almost 23.000 lines? No thanks.

AI wanting to include a reference to an issue without the right link and the context that it should be managed by renovate.

AI complaining about how another AI is pinning a version.

AI suggesting to add a test to verify that a function that has been already tested works, but doesn’t suggest to add E2E tests which was the right thing to do but wasn’t suggested in the issue but in the comments. Lack of context?

AI creating a PR with basically the same content of the PR on the left, by one of the maintainers, only 4 hours later. Did the AI know about the first PR and just used it without telling the contributor? Or did the contributor not research first if their issue was already solved before submitting their own solution?

Also: why are there so many spaces?

Now these examples may have inspired a few laughs. But some Pull Requests are so good, until we find unexpected behavior in testing, and the author can't explain what's happening because they didn't write the code.

Even more disappointing has been the surfacing of fake applications in the LFX mentorship program. This initiative exists to empower actual human beings, providing them with the tools they need to grow their expertise and climb the contributor ladder. Instead we've wasted time interacting with applicants who were wholly unqualified.

Read on if you're human

If you would like to contribute to CloudNativePG, here are the ways to do so:

Top comments (0)