Responsible Agentic Development

This post is mainly focused on agentic development via Claude Code or Codex. Recently, I've been getting lots of merge requests on my projects, which I am very happy for! I'm glad that people find my projects to be useful, fun, interesting, or some combination of the above. However, I've noticed that a lot of these merge requests have either been clearly AI-generated with little human-intervention.

I'm not against agentic development, and I actually support it. I feel that Claude Code and Codex are now very mature tools that are so helpful for development. I use them on a daily basis, and you can see me even pushing commits using them to fix small bugs. I want you to use it too, except, I want you to at least put some effort into reviewing your changes before sending them off to me! At the end of the day, the goal is that your changes do not look like an AI even generated them. There's too much "AI slop code" these days that make it past review, and even sometimes I find myself accepting some of these slop pull requests because I feel like it would be easier to fix them myself.

In general, Claude is very good at following existing codebase design patterns. Most of my projects are very big (tens of thousands of lines of code), and have been mature for a few years now, so it's not as big as an issue. However, Claude still slips sometimes. For example, I always add a newline at the end of all of my files, and sometimes it forgets to do that! And I get it, it's hard to pick-up these things especially if it's your first contribution. But I want you to try your best, so it will be easier for both of us – you will get your request approved faster, and I will be able to comment less nitpicks 😄.

Thanks for reading my post!