TraviaTechPie Review

The Story

On August 17, Cursor — the AI coding editor made by Anysphere, which SpaceX had just acquired for $60 billion — quietly shipped something that isn’t an editor at all. It’s called “Origin,” and it’s a code hosting platform. Repositories, pull requests, code review, browsing, merges, CI hooks. If that list sounds familiar, it should: it’s basically the job GitHub and GitLab have owned for the better part of two decades.

So the obvious headline writes itself. Cursor built a GitHub rival. And yes, technically, it did. But that framing misses the interesting part, and the timing gives it away. Origin launched on the same day GitHub was hit by a major multi-hour outage. You couldn’t script a more on-the-nose entrance if you tried. Except the outage is a coincidence that happens to point at the real story — which is not “here’s another place to store your code.”

Here’s the tagline Cursor is using: “Git hosting, at agent scale.” That phrase is doing a lot of work, so let’s unpack it.

For twenty years, code hosting has been built around a specific assumption: that a human being is on the other end. A person opens a branch, writes some commits over an afternoon, opens a pull request, waits for a teammate to review it, addresses comments, and merges. The whole ceremony — the review queue, the PR discussion thread, the merge button — is paced to human hands and human attention. GitHub’s entire design language assumes the bottleneck is people.

Now watch what happens when the person is replaced by an agent. Cursor has described Origin workloads at a scale that just doesn’t map onto that assumption: roughly 296,000 clones per hour, and up to 22.6 commits per second on a single repository. Those figures come from a demo Cursor showed at its Compile conference in June — vendor numbers, not audited benchmarks — so discount them accordingly. But even discounted, the point stands. An agent doesn’t write a few commits over an afternoon. It fires off dozens per minute, spins up and tears down clones constantly, and opens pull requests faster than any review queue built for humans was ever meant to absorb.

This is what “agent scale” actually means. It’s not a marketing flex about being fast. It’s a claim that the shape of the workload has changed — and that Git hosting designed for human cadence starts to strain when the traffic is machine cadence instead.

Origin is Cursor’s answer to that. And crucially, it’s plumbed straight into the editor. You reach Origin through a new tab inside the Cursor desktop client and a dedicated command-line tool, not a separate website you tab over to. There’s a one-click path to copy a GitHub repo into Origin, and it keeps syncing upstream changes so you’re not forced to abandon GitHub cold turkey. For now it’s an early beta, available only to paid Cursor subscribers.

That “plumbed into the editor” detail is the whole game. Cursor already owns the place where the code gets written and where the agent runs. Now it’s reaching down one more layer to own where the code lives. Editor, agent, and repository become a single loop, controlled end to end by one company. In platform terms, that’s vertical integration — and the prize is the control point over the developer’s workflow, the piece GitHub has held since Microsoft bought it.

And it’s worth being precise about who “one company” now is. Anysphere’s $60 billion sale to SpaceX closed on August 14 — three days before Origin shipped — folding Cursor into a SpaceXAI division alongside xAI. So the entity reaching down to own the repository layer isn’t just an editor startup; it’s a company that also builds Grok and needs exactly what Cursor produces: a direct channel to developers and a firehose of real-world coding data to train on. Owning where the code lives isn’t only a workflow play. It’s a data pipeline.

I covered a related idea a few weeks back, when researchers found that the real key to making coding agents cooperate wasn’t chat — it was Git itself, used as a coordination layer. Origin is that insight turned into infrastructure. If Git is how agents coordinate, then whoever hosts the Git repo sits at the center of the whole agentic workflow. Cursor clearly did the same math.

The Takeaway

Let me be balanced here, because the “GitHub killer” narrative is too easy and mostly wrong.

GitHub’s moat isn’t the git server. Anyone can run a git server. The moat is everything piled on top: GitHub Actions and the CI ecosystem, hundreds of millions of developers, the identity layer where your open-source reputation lives, and the fact that basically every tool in the industry already integrates with it. Origin, on day one, is an early beta gated behind a paid Cursor plan. That’s not a threat to GitHub’s install base. It’s a bet on a specific slice of the future.

And that bet is the part worth watching. Cursor isn’t really arguing that developers will abandon GitHub. It’s arguing that a new category of traffic — agents writing code autonomously — needs infrastructure built for it, and that the company owning the editor and the agent is the natural one to provide it. Inside Cursor’s own product, agents increasingly open pull requests on their own rather than a human doing it — and that share is the tell. If it keeps climbing, the question stops being “which website hosts my code” and becomes “which system can absorb an agent that never sleeps.”

This connects to a thread I’ve been pulling on for a while now. I wrote about a startup that raised money to let AI agents run an entire company through one API, and about Liquid AI shipping models small and efficient enough to run agents everywhere. The pattern underneath all of it is the same: as agents move from novelty to workhorse, the infrastructure assumptions we inherited from the human era quietly stop fitting. Identity systems built for people don’t know how to name an agent. Company tooling built for employees doesn’t know how to manage a bot. And now code hosting built for human review cadence doesn’t know what to do with 22 commits a second.

Origin might not win. Betas fail, and GitHub is enormous. But the specific claim underneath it — that agent-scale software development needs its own foundation, not a retrofit of the human-scale one — is going to keep showing up, in one product after another, whether or not this particular one sticks. That’s the real signal here. The GitHub outage on launch day was just dramatic timing. The real news is that agent-scale hosting shipped at all — and that it now sits inside SpaceX’s AI empire, pointed straight at the developer workflow Microsoft has owned for years.

Photo: Ilya Pavlov / Unsplash