-
Notifications
You must be signed in to change notification settings - Fork 28
Description
What problem does this address?
As a project all about experimenting with and promoting proper AI usage, it's to be expected that contributors are themselves using AI at various levels.
At the same time, effort involved in reviewing LLM-driven contributions are much higher than typical PRs, as they present a surface-deep picture of completeness and intentionality, and put the onus on code reviewers and maintainers to discern intent. Especially in a community-owned project, where we must make space for subjective preferences while enforcing high levels of code quality and maintainability, knowing whether something was engineered with consideration or statistically autogenerated is crucial to performing a thorough and efficient code review.
What is your proposed solution?
We add a checklist to our PR template, asking contributors to indicate if and how AI was used for this PR.
The goal is not to discourage AI usage, just to evaluate and understand to what extent it was used: e.g inline autocompletions are less likely to be hiding slop than limited chat copypasta than asking an agent to "clean up" your draft pr than vibe coding etc (the order here isnt important, just that there is one).
Bonus: We also get some fun data about who/how people are using AI to contribute, which is probably super helpful for making informed decisions, and guiding parallel discussions happening on different WordPress teams.
Something along the lines of :
🤖 AI Usage Disclosure
Please indicate if and how AI tools were used in preparing this Pull Request.
This helps reviewers understand context and intent, and supports ongoing discussions about responsible AI usage.
- No AI tools were used
- Minor inline autocompletions (e.g. IDE/code editor suggestions)
- Limited AI assistance (e.g. copy/paste from chat, small code/doc snippets)
- AI-assisted editing (e.g. asking an agent to clean up/refactor draft code or docs)
- Heavily AI-generated (e.g. large sections of code, documentation, or design produced by an AI tool)
- Other (please describe): ______________________
PS: The above example was 100% autogenerated by an LLM (GPT-5-mini via GitHub Copilot). How much easier does it feel to review and give feedback on, knowing I'm not particularly invested in it I just liked it better than the others I got out of my negligible prompting.
Metadata
Metadata
Assignees
Labels
Type
Projects
Status