On the morning of August 18, a new tab appeared inside the Cursor editor: Codebase. After weeks of quiet, Cursor began rolling out Origin, its own code hosting platform, to paying users. Roughly three and a half hours later, GitHub's status page registered a global outage lasting 6 hours and 42 minutes, with web and API error rates near 20% and archive/raw file download errors climbing to 50%. A carefully staged product launch collided with an uncontrolled failure at the incumbent giant—on the same working day.
Origin is not another GitHub. Its goal is to put code, pull requests, and AI agents in the same window. Developers can name repositories with dedicated URLs, push via the command line, and rely on built-in hosting capabilities—storage, permissions, checks, and merging—without ever opening a browser. Agents act directly on the files on screen: editing PRs in place, pushing branches, all inside the editor where code is written. On launch day, Origin shipped with three partners: Vercel generates preview deployments for every PR and publishes to production on merge; Depot and Buildkite provide continuous integration, and both can run existing GitHub Actions workflows unchanged.
The sharpest move, however, is that Origin lets GitHub keep serving as the "data source." Once a GitHub organization is connected, existing repositories appear alongside Origin-native repos, pushes still sync to GitHub, read/write permissions mirror GitHub's existing settings, and PR conversations sync bidirectionally—comments on one side appear on the other within seconds. This read-only-mirror approach sidesteps the high risk of source-hosting migrations, which involve CI, compliance evidence, audit records, branch protection rules, and every engineer's habits—almost no team approves a full migration for an early beta. But if Origin's review experience is good enough, developers' actual working center of gravity shifts, and the data source eventually follows attention.
The timing collision with GitHub's outage was widely read as opportunistic, but it's more accurate to say that agent-era, high-frequency commit/clone traffic is overwhelming traditional Git infrastructure. Origin's agent-grade performance figures: 296,000 clones per hour and 22 commits per second per repository. When a parallel agent workflow can generate hundreds or thousands of commits and pulls in a few hours, throughput and conflict resolution on traditional Git platforms become the bottleneck—while Origin builds parallel PR management, automatic merge conflict resolution, and 400ms global latency in as native capabilities.
Origin's core pitch: agents in the same interface as code and PRs
Cursor didn't reinvent version control—it reinvented the interface between agents and version control. Developers can ask questions about on-screen files, hand review feedback to an agent to revise a PR in place, or have it push branches directly. Workflows that previously bounced between the GitHub web UI, CI logs, code review comments, and the IDE are now compressed into one.
The "read-only mirror" is the decisive move in the migration war
The real cost of migration isn't tooling—it's compliance evidence, audit records, branch protection rules, and the habits of every engineer. By syncing to GitHub, inheriting GitHub permissions, and bidirectionally syncing PRs, Origin effectively outsources migration cost to attention migration: once developers are used to viewing diffs, commenting, and merging in Origin, the data source follows. It echoes how GitHub marginalized SourceForge with "social coding"—except this time the social surface is agents.
Agent-era Git infrastructure bottlenecks are now on the table
GitHub's 6h42min global outage won't be the last. With parallel agent workflows committing and cloning at tens of times the historical rate, throughput, concurrent merges, and Actions queueing on traditional Git platforms are being crushed. Origin's hourly clone count (296,000) and per-second commit rate (22 per repo) are SLAs traditional hosting never promised—not a gimmick, but an engineering baseline for the agent era.
What changes
Origin redraws the boundary of AI coding tools: Cursor moves from IDE to platform, turning "writing code" into "operating code." When agents take over code hosting, GitHub loses not just engagement time but the entry point to developer workflows in the AI era. And GitHub's global outage that day put the fragility of single-platform dependency in front of every CTO—engineering risk has never felt this concrete.
*Note: This post is a forum discussion on zhichai.net analyzing the launch; figures cited are from the original post.*