Cursor Origin: The Git Forge Built for AI Agents

Cursor announced Origin, a Git forge rebuilt for AI agents, at its Compile conference on June 16. Origin addresses GitHub's inability to handle AI agent workloads, such as 275 million weekly commits and 296,000 clones per hour, with parallel-first merging, automated conflict resolution, and MCP extensibility. The product ships this fall and positions Cursor as a full-stack development platform competing with GitHub.

Cursor announced Origin at its Compile conference last week – a Git forge rebuilt from scratch for AI agents. The pitch is blunt: GitHub was designed for humans pushing code a handful of times a day. AI agents are now committing 22 times per second to a single repo, cloning projects 296,000 times an hour, and generating parallel merge conflicts at a rate that no human engineer can resolve. Origin is built to handle that. GitHub was not. GitHub Was Not Designed for This The numbers are hard to ignore. GitHub is now processing 275 million AI agent commits per week – roughly 14 times the total commit volume of all of 2025 combined. In March 2026 alone, AI agents opened 17 million pull requests. The platform suffered nine outages in May and has been running below its 99.9% enterprise SLA. The infrastructure response? GitHub moved onto AWS https://byteiota.com/github-ai-infrastructure-crisis/ – not Azure, its owner’s own cloud – because Microsoft’s infrastructure could not absorb the demand. GitHub was built in 2008. Its data model assumes humans are the primary actors: commits arrive at human speed, pull requests are reviewed by engineers in browsers, and merge conflicts get resolved manually after reading a diff. None of those assumptions hold when agents are the ones doing the work. What Origin Actually Changes Cursor unveiled Origin at its Compile conference https://cursor.com/compile on June 16. The product is built on re-architected technology from Graphite, the code review startup Cursor acquired in December 2025 for more than its $290M valuation. The Graphite team put it plainly in their acquisition post https://cursor.com/blog/graphite : GitHub was built for a world where only humans commit code. The thesis of Origin is that the hosting layer needs to be rethought from scratch for the agentic era. Three things distinguish Origin architecturally from every existing Git host: Parallel-first merge strategy. Origin assumes multiple agents are editing overlapping code simultaneously. It does not treat this as an edge case – it treats it as the default. Automated conflict resolution. When agents push conflicting branches across dozens of files, there is no time for a human to step in and resolve them manually. Origin handles this with AI-powered conflict resolution built into the merge layer. MCP and API extensibility. Review state is machine-readable and machine-actionable from the start. Agents can drive the forge programmatically via Model Context Protocol, not just through a browser UI built for humans. The storage architecture backs this up: NVMe-backed Git fileservers for speed, S3 as the source of truth, and what Cursor calls infinite replicas for global sync and failover. The demo numbers were 296,000 clones per hour and sub-400ms global synchronization latency. The Full-Stack Play Origin is not a standalone product – it is the third layer in Cursor’s broader bet. Cursor Composer handles the editing layer, where agents write code. The Graphite acquisition covers the review layer, with stacked PRs, AI code review, and merge queues. Origin closes the loop at the hosting layer. If this works, Cursor is no longer an editor. It becomes the control plane for the entire software development lifecycle: write, review, and host, with AI agents threading through all three. That is a meaningful shift in where the leverage sits in the toolchain. It also puts real competitive pressure on GitHub https://thenewstack.io/cursor-origin-github-disruption/ for the first time in years – not from GitLab or Bitbucket, but from a company that came at software development from the AI-first direction. What You Should Do Now Origin ships this fall – no specific date, no pricing announced. The waitlist is open at cursor.com/origin https://cursor.com/origin . If you are running parallel AI agents today and hitting GitHub scaling limits, Origin is the most credible near-term alternative being built. If you are not, the announcement is still worth understanding: the infrastructure layer of software development has not been seriously contested since GitHub became dominant. That is changing. The product is not here yet. The problem it is solving is.