Session Description

LLMs have lowered the barrier to contributing to open source, but for maintainers the picture is more complicated than “more contributions = better”. This talk looks at how the maintenance of open source projects, including the Scala 3 compiler, changes when AI-generated issues, comments, PRs and even design proposals start showing up.

We will walk through a real-looking low-effort LLM contribution and compare it with an actual fix. The interesting part is not whether the code compiles, but whether it solves the real compiler bug – or merely patches the single reproduction case that was given to the model.

From there, we will zoom out to the broader stream of AI-generated open source activity: confident but wrong explanations, noisy GitHub comments, suspicious proposals, bot-authored pull requests. These examples show how much of maintainer work is not just writing code, but understanding intent, verifying claims, and deciding what is worth trusting.

This is not a talk about mocking AI users. Many of us use LLMs every day, sometimes very effectively. The goal is to show what maintainers now have to review, verify, and make sense of – and why it is surprisingly hard to tell whether the LLM era is making open source maintenance easier, harder, or simply different.