Cursor's Merrill Lutsky on Origin — rebuilding Git infrastructure from scratch for agentic coding at SpaceX scale

Merrill Lutsky built Graphite to add stacked diffs to GitHub. Now, via Cursor's acquisition by SpaceX, he's rebuilding Git infrastructure from scratch.

The product is Origin, Graphite's new Git layer designed specifically for agentic coding at scale. The core argument is that every piece of existing software development infrastructure was built for humans writing code, and that foundation is crumbling under agentic load. Teams that once pushed tens or hundreds of times a day are now pushing thousands of times per hour, and current Git tooling wasn't designed for that throughput.

Origin is built to handle it. Lutsky says internal simulations are running at 80 clones and 22 pushes per second with no downtime. The goal isn't just raw performance — it's owning enough of the infrastructure layer to embed agents more deeply into the full pull request lifecycle. Today, an agent writes a PR and then disappears the moment it's opened. Lutsky wants agents that persist through review comments, CI failures, and merge conflicts, moving toward what he calls "full self-driving PRs" that can reach production without human intervention.

“Every single piece of the software development life cycle that we have today was all designed for a world where humans write every line of code... We set a goal for end of July for wait list sign ups that we wanted to hit, and we hit that in the first twenty four hours since we announced the product.”

Waitlist

The acquisition hadn't closed as of the conversation, but commercial momentum is already visible. Graphite had set an end-of-July waitlist target for Origin signups and hit it within 24 hours of the announcement, following the product's debut at Cursor's first developer conference, Compile, in San Francisco.

The harder lesson from Graphite

Lutsky is candid about what building Graphite taught him. The product was always constrained by the platform it sat on — availability issues, limited UI extensibility, and infrastructure downtime that pulled Graphite down with it. He notes the irony that companies are now asking AI agents to clone Graphite's pull request UI, only to quickly discover why it was hard to build. Origin is, in part, the answer to the problem Graphite never fully solved: owning the infrastructure rather than renting it.

On code quality, Lutsky is clear that the future isn't 10,000-line PRs to change a button color. But the throughput surge is real, and the solution he's betting on is more agentic review and autonomous refinement before a human ever looks at the diff — not a reversion to sparse, human-paced commits.