I Am Not a Reverse Centaur
Open‑source maintainers are increasingly overwhelmed by low‑quality, AI‑generated pull requests, prompting calls for more friction—such as requiring issues before PRs—and even blanket rejection of contributions that show no clear human involvement. Commenters are split between frustration at “drive‑by” LLM slop and excitement that non‑programmers can now build custom software, raising deeper questions about what counts as real contribution, whether open source still “matters,” and how to preserve trust, craft, and sustainability in a world of easy code generation.
Impact of LLMs on Open Source and PR Quality
- Many maintainers report a flood of low‑effort, AI‑generated pull requests that waste review time.
- AI makes it easy to generate “plausible” but shallow code and write verbose, generic PR descriptions, breaking the old norm that authors expend more effort than reviewers.
- Some see LLM‑generated libraries and patches as adding noise, making it harder to find and trust high‑quality projects.
- Others note AI can still produce useful fixes, especially when paired with good tests, but stress that unreviewed AI code is not a drop‑in replacement for well‑maintained libraries.
Maintainer Responses and Process Changes
- Strong support for adding friction: require issues (or even discussions) before PRs, auto‑reject PRs without linked issues, and prioritize clear human explanations of problems.
- Some argue this friction filters out “drive‑by” contributions and bikeshedding; others find it demotivating or bureaucratic.
- There is tension between curating a small, respectful contributor set and maintaining the “open” feel of open source.
Empowering Non‑Programmers vs “Reverse Centaur” Concerns
- Several commenters celebrate that non‑coders can now create custom tools and “home‑cooked apps,” especially for niche or accessibility needs.
- Critics compare mass LLM use to unskilled people building houses with power tools: empowering but risky for shared, long‑lived software.
- “Reverse centaur” is framed as humans mechanically executing AI plans; some see ticket‑driven corporate work as a precursor to this.
Pride, Craft, and Use of Tools
- One camp insists pride should correlate with effort, skill, and understanding; mere prompting or commissioning work is not a meaningful “accomplishment.”
- Another camp argues it’s valid to feel proud of causing a useful thing to exist, even if tools (compilers, libraries, LLMs) do most of the mechanical work.
Does Open Source Still Matter?
- Some fear OSS is being strip‑mined to train models and that open licensing mainly benefits AI labs.
- Others counter that modern software (including AI stacks) still critically depends on open source, so it remains essential, though under strain.
- A few suggest “free software” ideals (copyleft, anti‑relicensing, user freedoms) may now matter more than open source as a branding or business model.
Detection and Use of AI in the Workflow
- Detecting AI‑generated code is mostly “by vibe” (style, boilerplate, tone) and considered unreliable to automate.
- Some propose using LLMs themselves to pre‑triage or summarize PRs, but there’s discomfort with leaning further on the same technology causing the problem.