A year after declaring LLM‑generated code unusable for his own work, the open‑source maintainer now faces a flood of pull requests produced by AI. Most contributors submit changes without human oversight, prompting him to label the workload a reverse centaur problem, a phrase borrowed from Cory Doctorow describing developers puppeted by machines. He notes few patches help, but most bring vague LLM changes.

To curb the noise, he tightened contribution guidelines: contributors must first open an issue, discuss the change, and then submit a pull request. This front‑loading lets him verify a real person is behind the proposal. When a contributor follows the issue workflow, the maintainer can assess intent quickly and merge confidently. PRs lacking human proof are closed, sparing him hours of reviewing low‑value AI output.

He admits the stricter process may discard useful fixes, but prefers protecting his projects from noisy AI edits. He urges developers who rely on LLMs to describe problems in issues rather than token‑heavy PRs, and even suggests donations to prioritize their tickets. The maintainer will continue enforcing these rules to keep his open source work manageable.