Several prominent open-source teams are changing how they accept contributions: instead of reviewing outside pull requests, they increasingly turn community reports into issues or discussions and ask their own AI agents to produce the code. Flue and the source-available drawing tool tldraw automatically close external pull requests, according to Latent Space.
Vercel has taken a less absolute but highly automated route for its AI SDK, which receives more than 20 million npm downloads a week. Its internal “software factory” assigns separate agents to reproduce bugs, implement fixes and review changes. Vercel says the system now authors 25% to 35% of merged pull requests and closes 70% to 80% of issues after four weeks in operation.
Astro has also deployed automated triage after years of issues arriving faster than maintainers could handle them. The shift may reduce the burden of reviewing low-quality, AI-generated patches, but it removes a traditional route through which contributors learn and eventually become maintainers. The emerging compromise keeps reporting, discussion and project judgment open to the community while maintainers retain control of code generation and review.