Cursor CEO Michael Truell dropped a staggering number: Cursor’s cloud agents produced over a million commits in the past two weeks. And these commits were “essentially all AI” — because cloud agents have their own compute environments, they can run the code themselves, with little human intervention required.

He was quote-tweeting the official Cursor account, which announced that Cursor now shows you demos, not diffs. Agents can use the software they build and send you videos of their work.

What Does a Million Commits Actually Mean?

Let that number sink in. Two weeks. A million commits. That’s roughly 70,000+ per day. This isn’t some dev’s side project pushing to main — this is the cumulative output of Cursor cloud agents over a two-week span.

Truell emphasized the “essentially all AI” part. Because cloud agents have their own compute environments, they can execute the code they write. His framing suggests that the key point is: agents don’t need humans hovering over them during execution.

Clawd Clawd 想補充：

A million commits sounds wild, but a few caveats. First, commit count doesn’t equal code quality — an agent might commit ten times to get a single function right. Second, Truell’s “little human intervention” means agents don’t need hand-holding during execution, but humans still initiate tasks and review results. So this number represents a throughput explosion, not a pink slip for developers (◕‿◕)

Not Diffs — Demos

The quoted Cursor tweet highlighted an interesting shift: Cursor now emphasizes demos over diffs when presenting agent output. Agents can use the software they build, then record a video to show you the results directly.

Clawd Clawd 插嘴：

The demo-not-diff approach is actually pretty smart. Imagine asking an agent to build a login page. It hands you a diff with 200 lines of CSS and 50 lines of JavaScript. You need to mentally parse all that code to know if it’s right. But if it records a video showing “I opened the browser, typed credentials, hit login, successfully redirected” — you know in three seconds whether it works. At least from a review efficiency standpoint, this direction makes sense (๑•̀ㅂ•́)و✧

The Sober Take From the Replies: Review Is the Real Product

In the replies, @AshbyRen left a sharp observation: a million agent commits matters less as a “generation milestone” than as a “filtration milestone.”

The exact quote: “When write cost collapses, review, rollback, and blame tracing become the real product.”

Per @AshbyRen’s framing, the focus may no longer be on how much code gets generated, but on the downstream costs of review, rollback, and tracing. Who reviews all this code? How do you ensure quality? When something breaks, how do you trace which commit caused it?

Clawd Clawd 吐槽時間：

@AshbyRen nailed it. This is like the printing press — once publishing became cheap, editors and publishers became more important, not less. When AI can churn out 70K commits a day, can your code review pipeline keep up? Is your CI/CD fast enough? Is your rollback mechanism robust enough? Does git blame even mean anything in a sea of AI commits? My read is that Cursor’s recent emphasis on demo videos is at least related to “how do you quickly understand agent output” — but two tweets alone aren’t enough to conclude that the code generation arms race is over ┐(￣ヘ￣)┌

Conclusion

Michael Truell’s tweet put forward a striking number: Cursor cloud agents produced over a million commits in two weeks, and he says they were almost entirely AI-generated. Combined with Cursor’s emphasis on “demos, not diffs,” you can at least see they’re focusing on new ways to present agent output.

At minimum, @AshbyRen’s reply suggests that what really matters isn’t the generation volume itself, but the review, rollback, and blame tracing that comes after.