Cursor's Origin gets a real number: 22.6 commits/sec in a live demo

A well-followed dev account just put a real number on Cursor’s Origin, the code platform built for AI agents instead of human coders. In a conference demo, a single repo hit 22.6 commits per second. That’s the first concrete stat we’ve seen since Origin launched, and it’s worth pausing on because of what it actually reveals about the design.

GitHub was built around a simple model: one author writes code, another human reviews it, changes land one at a time. Origin flips that. The person who posted the number framed it plainly as “GitHub for humans, one author, sequential review, versus Origin for AI agents as primary users.” When agents are the main users instead of people, you don’t get a handful of commits an hour. You get dozens per second, running in parallel, because the whole point is that many agents are working the same repo at once.

Here’s why this matters if you run a company that ships software, even a small in-house tool or a client project. Your current review process, your approval chain, your sense of what “a day’s work” looks like in a codebase, all of that assumes human pace. If your dev team starts running agents inside a tool like Origin, the volume of change coming at you will outrun any manual review process almost immediately. That’s not a future problem. It’s a this-year problem for anyone letting agents write and commit code directly.

The number itself came from one demo, so treat it as a snapshot, not a guaranteed baseline. But the direction is the real story: tooling is now being built around the assumption that AI, not people, is the primary user. Keeping a clear log of who or what changed what, and why, is exactly the kind of thing we build into an AI command centre.