RIP Pull Requests (2005–2026)
Site Owner
Published on 2026-04-30
GitHub just allowed repos to disable Pull Requests for the first time in history. In 21 years, the PR went from invention to industry default to potential extinction. Here's the technical, cultural, and economic story of why AI is about to kill the most fundamental workflow in modern software development — and what comes next.
RIP Pull Requests (2005–2026)
GitHub just did something unthinkable. For the first time in history, open-source repos can now disable Pull Requests entirely.
21 years. That's how long PRs lived.
The Most Boring Revolution in Tech History
Let's be honest: nobody ever wrote a TED talk about Pull Requests. There were no startup founders bragging about their PR workflow at Demo Day. No VC firm published a thesis titled "Why Pull Requests Will Change the World."
And yet.
Pull Requests — invented in 2005, popularized by GitHub in 2008 — quietly became one of the most consequential coordination mechanisms in all of software. They were the atomic unit of how billions of dollars of open-source work got merged. How code review happened. How trust was negotiated between strangers on the internet. How the modern internet got built.
You used git checkout -b feature/something, wrote some code, opened a PR, and the world sorted itself out.
It was boring in the way that plumbing is boring. Until it wasn't there anymore.
The Death Sequence
The warning signs have been accumulating for a while. Pete Steinberger — well-known in developer tooling circles — has been saying for months that he no longer wants Pull Requests. He wants Prompt Requests.
The logic is brutally practical:
- No merge conflicts. A PR is a diff against main. Two people touch the same file, you get a conflict. A prompt is a description of intent — no file-level collisions.
- Easier for maintainers. If you don't like what the AI did, you fix the prompt, not the code. The delta is a conversation, not a patch.
- No supply chain attacks hiding in "innocent" diffs. This one matters. PRs are a legitimate vector for malicious code. A prompt requesting "improve the auth module" is a lot harder to weaponize than a 300-line diff with a subtle backdoor buried in line 247.
The PR model was built for human collaboration. It assumes both parties are humans, both have context, both are negotiating in good faith. One of those assumptions is already broken.