Lauren says she shipped 2,500 PRs to production last month
The talk was meant for Cursor Compile in London, and she put it on X instead.
lauren (@poteto) has posted the walkthrough of how she says she shipped 2,500 PRs to production last month. The talk was originally prepared for Cursor Compile in London, which she missed because she was livestreaming for Grok, so she put the whole thing on X for free instead. Her only viewing note is practical, watch it at 2x speed, because she talks slowly.
here's how i shipped 2,500 PRs last month to production
The number is the part that will get argued about, and it is worth being precise about what it is. It is her own count of her own output, posted by her, with no independent verification attached. A PR is also not a fixed unit of work. Two thousand five hundred merged pull requests in a month averages out to more than 80 a day, which tells you immediately that most of them cannot be large, and that whatever she is describing is a pipeline rather than a person opening tabs.
What the post does not do is summarize itself. There is no thread breaking down the setup, no list of tools, no before and after numbers on review time or revert rate. The claim is in the tweet and the method is in the video, which means anyone who wants to evaluate it has to actually watch it rather than react to the headline figure.
For working engineers
The interesting question is not whether 2,500 is impressive, it is what the review and revert story looks like at that volume, and that is exactly the part a raw PR count hides. Engineers in this publication's earlier coverage have been circling the same problem from the other direction, from Gary Bernhardt cutting most agent diffs to a quarter of their size to Mitchell Hashimoto's rule that you explain the code or you do not ship it. Throughput claims and review discipline are the two halves of the same argument, and right now one of them has a number attached and the other does not.