Correction, October 13, 2026: Flue and tldraw restrict external pull requests. Vercel and Astro use maintenance agents without the same blanket restriction. The issue-closure figure describes the share of closed issues handled by Vercel’s factory.

A pull request is a proposed code change sent to a project’s maintainers. For 18 years, it gave outside contributors a path to have their work reviewed and merged. Some active AI-focused projects are now closing that path.

“But now some of the top AI-native open source projects are shutting PRs off,” Richard MacManus wrote at Latent Space. Those projects use their own AI agents to write, review, and merge more changes. Human contributors get routed toward bug reports and discussion instead.

What he found

MacManus highlighted Vercel’s AI SDK. The open source toolkit gets more than 20 million weekly npm downloads, according to the figures in his post. By late June, it had more than 1,000 open issues and almost 800 pending pull requests.

Vercel responded with what it calls a “software factory.” Separate agents classify issues, reproduce bugs, write fixes, and review changes. A human still makes the final merge decision.

Vercel reports that its factory authors 25% to 35% of weekly merged pull requests. In July, it handled over 75% of closed issues. Open bugs fell roughly 25% from late June to early August.

Astro creator Fred Schott described a similar problem. “For five years, we were in this place where issues came in faster than we could handle them,” he said. Astro uses agents for triage and reproduction before maintainers review a fix.

MacManus distinguishes that automation from projects that restrict external pull requests. Flue, created by Schott, converts outside pull requests into issues or discussions. The source-available drawing tool tldraw also auto-closes external pull requests. Founder Steve Ruiz called the decision a response to changes in coding and code security. He said outside contributions should focus on reporting, discussion, perspective, and care.

What MacManus gets right

MacManus identifies a real change in open source maintenance. Agents can now handle repetitive work that once consumed maintainer time. Projects can process more reports without asking maintainers to write every fix themselves.

The tradeoff is less clear. Pull request review was also a way to discover future maintainers. It taught contributors how a project worked. Closing that channel removes a path into open source leadership.

The examples also come from well-funded projects. Their systems may not fit smaller projects without the budget to build and supervise several specialized agents. The trend is important, but its reach remains uncertain.

What it means for builders

Maintainers facing a large issue queue can separate the work into stages. One agent can classify reports. Another can reproduce the problem. A third can propose a fix. Keep a human responsible for the final merge.

Contributors should read a project’s contribution rules before writing code. Some AI-native projects may close a pull request automatically. Reporting a problem or explaining a user need may now be the accepted way to help.

End of article