Jakub Jankiewicz argues that AI-generated code should be open source or placed in the public domain. His case rests on the collective work used to train models and the uncertain copyright status of output with little human authorship.

He acknowledges that software often combines human and machine contributions, but argues that code produced with little more than a prompt should be shared. This is a proposal about what developers should do, not a statement that current law requires publication.

The 310-comment thread on Hacker News complicates both parts of the argument.

What the thread adds

  • blagie describes an alleged case where a company replaced an open-source system with an AI-generated clone; Aurornis asks for concrete examples. The allegation is not independently verified here.
  • jrmg cites US Copyright Office guidance distinguishing purely generated material from human-authored expression and calling for case-by-case evaluation of human contributions.
  • glitchc questions how a copyright license can govern material that has no copyright owner.
  • anself describes substantial human testing and design around agent-generated code, and objects to mandatory publication of private projects.
  • x3haloed favors using AI to produce free software for everyone, but questions requiring all model output to be open source.

Handles are pseudonymous, HN provides no per-comment scores, and this is a slice ordered by HN’s ranking, not a consensus.