Zed killed pull requests on its own repo, then shipped Delta

Zed turned off pull requests on its Delta repository. Since then, 33 of its developers have landed 570 changes on main through Delta threads. On September 16 the company opened Delta itself as a public beta: a shared environment where developers and coding agents write, review and revise code together, with no pull request involved.

CEO Nathan Sobo is blunt about the target. “Everyone is in a race to replace GitHub right now,” he wrote in a blog post on Wednesday. Zed itself is only “a few months away” from leaving GitHub for its internal work, Sobo says, per The New Stack.

GitHub’s load problem shows up in the figures. GitHub logged hundreds of incidents in the 12 months leading into June. Over the same period, commit volume went from around 1 billion across the whole of 2025 to 1.4 billion a month by April, and GitHub put it at 2.9 billion a month by August. That load was behind a near-eight-hour outage in August, with web and API error rates peaking at around 20%.

Sobo’s argument against the pull request is about review. “Since GitHub introduced pull requests over 15 years ago, they’ve become the standard way to ask teammates to review changes to your codebase,” he writes. “But with agents generating so much code, the diffs we’re asking each other to review have mushroomed.” And when much of the reasoning behind a change happens in a conversation with an agent, a pull request gives the reviewer the resulting diff while the conversation that produced it stays somewhere else.

Delta’s answer is the thread. A thread is a running record of one agent-assisted task, and it keeps the conversation and the changed files connected. You give an agent a job, keep discussing and refining it, and later share the whole body of work with a colleague. Each thread can run against its own copy of the project, so several tasks move forward independently without every agent editing the same checked-out files. A reviewer can join the existing thread or open a separate review thread, question the agent that wrote the change, try revisions, and feed accepted changes back into the original work. Sobo argues this lets the next person pick up where the previous developer stopped, with the same agent and context, instead of piecing together the thinking behind a diff after the fact.

Underneath sits DeltaDB, which Zed first teased in August 2025 alongside a $32 million round led by Sequoia Capital. It is an operation-based version control system. Git waits for a developer to package work into a commit. DeltaDB instead records individual events as they happen, both code edits and activity in the conversation, and uses that history to keep everyone in the thread in sync. Zed calls each of those records a “delta.” Git stays in place for now: Delta works with ordinary Git repositories, branches, commits and remotes, and DeltaDB adds a layer of history between commits that Git would otherwise throw away.

Zed was founded in 2021 by veterans of GitHub’s Atom editor team, Sobo among them after nine years on Atom. It began as a fast multiplayer code editor written in Rust, and by 2025 AI editing and agent features had become central to it. Delta went into private beta in August, pitched as a place to code with agents, share ongoing threads and review changes with the agent’s context still attached.

Still tied to GitHub

“We now build and collaborate on Delta entirely within Delta,” Sobo writes. Getting off GitHub for good still takes work. The next big dependency is Git storage, which Zed plans to move into DeltaDB, and CI and releases have to move too. Sobo calls CI largely a solved problem and says Zed will integrate existing options rather than build its own just to replace GitHub. One item on Delta’s roadmap shows how close the two still are: repository-based access, which will use a GitHub repository’s existing permissions to decide who can see shared Delta threads. So for now, even the GitHub replacement asks GitHub who is allowed in.

The open-source Zed editor repository is a separate case. Its established community reports problems and proposes changes there, with hundreds of outside developers contributing each month. Those contributors can use Delta to expose the agent session behind their work and still submit the final change as a conventional pull request, so the Git workflow stays the same for anyone who does not use Delta.

“We’re moving more thoughtfully with Zed’s public repo because we have hundreds of monthly contributors who depend on that workflow,” Sobo told The New Stack. “GitHub has an established social component that will take longer to replace, and we’re not going to strand contributors to prove a point.”

Delta ships as a desktop app for macOS, Linux and Windows, plus a browser version for viewing, sharing and reviewing threads. It stays free for the whole public beta. Paid individual and team plans will follow, and Zed says there will always be a free version.

Thread versus commit

Zed is not the only company working on this. Cursor, now owned by SpaceX, formally launched Origin in August. Origin puts Git hosting, pull requests and Cursor’s coding agents in one product, while existing GitHub repositories can remain the source of truth. GitLab has its own “next-generation source-code management” project, Project Switch, in private beta. Origin keeps the pull request, and Delta is built on the bet that it goes away. “I believe the thread will replace the commit or branch as the fundamental unit of software development,” Sobo says. In his view commits remain useful checkpoints, but they lose everything that happens between them: the exchange with the agent, the decisions along the way, the small edits that add up to the final code. “Delta’s advantage is that we’re building around the thread from the start, including the infrastructure underneath it.”

He expects the market to settle the way the last one did. “Git became the common foundation because it was open and everyone could build on it,” Sobo says. “GitHub won the layer above through network effects. I expect lots of viable products will compete on the agent experience, with a common infrastructure underneath.”