I don't want your PRs anymore

2026-05-15 · source · source ↗ · raw capture · wiki page

Why I don't want to merge your PR

Three core concerns with outside PRs:

  1. Security risk — "I always have to assume that you might be trying to sneak in something malicious along with your changes"
  2. Subjective style differences — formatting, dependencies, and approach preferences vary between people
  3. Coordination overhead — review cycles, CI waits, merge conflicts, and timezone synchronization issues

With LLMs, these tradeoffs shift: no security worry about AI-generated code, style guidelines can be codified once, and iteration happens at the maintainer's own pace.

The nature of software development has shifted

Source code is "an intermediate formalized layer between ideas in the developer's head and instructions for the computer." The real bottlenecks now are:

"the code in your PR doesn't help me much with any of these"

How can you help instead

Six alternative forms of contribution:

  1. Give feedback — Users can report what works and what doesn't, since maintainers often lack time to use their own software thoroughly
  2. Discuss ideas — Sharing diverse perspectives helps shape what to build and how
  3. Report & investigate bugs — "A good bug report is 3/4 of the bug itself being fixed" — well-described reproduction steps are highly valued
  4. Prototype changes — Reference PRs and the prompts used to generate them are welcome. "A quick glance at code implementing something can still be helpful, even if I don't end up merging it." Sharing the "source" (prompt) lets the maintainer reuse and refine
  5. Review code & point out problems — Extra review eyes help since the author is bottlenecked on reviewing
  6. Fork & report back — "Just fork. Add support for your own use case, do things your way, ask neither for permission nor forgiveness." This saves the maintainer time on consensus-building and multi-use-case design

Central thesis

LLMs have made code generation cheap and low-risk for maintainers, while human-contributed PRs carry trust costs, style friction, and synchronization overhead that now outweigh their benefits. The author isn't rejecting help — they're redirecting it toward higher-leverage activities: feedback, design discussion, debugging, prototyping with prompts, code review, and independent forking.