Open source maintainer bans LLM-generated pull requests: "I am not a reverse centaur"

Miguel Grinberg, a veteran open source developer, says the flood of contributions to his projects has shifted almost entirely to LLM-generated code — and he refuses to spend his time cleaning up after machines. Borrowing Cory Doctorow’s term “reverse centaur” (a human puppeteered by a relentless machine rather than aided by one), he argues that reviewing drive-by AI pull requests would turn him into exactly that: a skilled engineer reduced to quality-control labor for code extruded by tools he chose not to use himself.

His response is a hard policy change. Where an unsolicited PR was once a welcome sign of genuine community effort, he now treats it as a red flag. Contributors must first open an issue and discuss the proposed change before writing code; PRs that arrive without prior discussion — or without evidence of a human behind them — are closed immediately, no questions asked. He acknowledges this may cost him the occasional useful fix, but argues that triaging slop to find it is no longer a justifiable use of his time. Users who rely on LLMs are told to simply describe their problem in an issue, in their own words, and let him do the work.

The piece closes on a bleaker note about open source itself: Grinberg still maintains his existing projects but has lost the urge to publish new ones, sensing declining interest in both open source and coding as a craft. The episode illustrates a growing structural problem for maintainers everywhere — AI tools have collapsed the cost of generating contributions while leaving the cost of reviewing them untouched, pushing that burden onto unpaid humans.