Post

I’ve noticed something slightly strange in one of my open-source projects: a contributor and I are both using LLMs to write our comments. It can be confusing, but the added context also makes the replies much better. Essentially, its agent2agent communication, with GitHub PR comments as the proxy.

But in general, PRs for open-source projects have gotten much better because as the description clearly outlines what the user wants to achieve, and the LLM adds context. That's better than an issue with a one-liner where we have to go back and forth a lot to figure out what the user actually wants.

There's definitely a middle ground to strive for here; I've read comments in issues and PRs, clearly written by LLMs, and they are so full of flowery language and tangential detail that they're pretty much incomprehensible, and as useless and frustrating as one-liners written by lazy humans.