Post

Marco Ieni
@ieni.dev

I changed the Pull Request template of release-plz. I realized I don't care if you used AI to write code, or how you used AI, or which model you used. I just want to make sure you reviewed the code before submitting it. 🧵

Marco Ieni
@ieni.dev

Generally, Astra will review your code first. If it says there's a better implementation I'll never review your PR. Instead, I'll let Astra + Fable 5.1 write a new PR. If Astra think it makes sense, Astra will include your commits in the new PR to acknowledge your work.

0 reposts0 likes

Marco Ieni
@ieni.dev

Almost all PRs I received in the last months contained AI generated code. I trust my prompts/harness/skills more then the ones of random external contributors. Why should I spend my time reviewing their AI generated code instead of mine?

Chraz
@edu.runlinea.com

The PR template change is the real tell. Once almost every external PR is model-written, asking which model they used stops being useful. Running their commits through your own harness, and only reviewing when that pass is clean, is how spare-time maintainership stays possible.

Marco Ieni
@ieni.dev

That said, I still require your human review before submitting a PR to avoid getting tens of low-quality PRs and wasting token reviewing them. I also require human review (ideally human voice) in issues, discussions, comments, etc.

Marco Ieni
@ieni.dev

Another thing I'm experimenting with: I created a bot account (marcoienibot). Agents are free to use it to create PRs with it when I ask to fix an issue. I'm also experimenting with "review -> fix" loops. Still needs improvements because sometimes PRs becomes too big.

Marco Ieni
@ieni.dev

This is the most sustainable way I found to keep maintaining release-plz in my spare time. It seems like it's working. In the last month: - 39 PRs merged in main excluding automations such as renovate. - closed 18 issues. Lots of bug fixes!