AI Code Hosting: From Human Pull Requests to Agent Workflows
AI code hosting is the redesign of code repository platforms so they can reliably store, index, and manage software created and reviewed at machine speed by AI agents instead of humans. That shift is forcing the industry to rethink what a “repo” is for. GitHub popularized pull requests for human review in 2008, but the model now strains under constant, automated activity. The New Stack reports GitHub is processing about 1.4 billion commits per month, with AI agents alone generating more than 17 million pull requests monthly, and the platform has logged hundreds of incidents in the last year. Former GitHub director Brian Douglas says “agents are quickly killing the will for doing open source” as developers move reviews into AI tools. Into this pressure step Cursor’s Origin, GitLab’s evolving AI stack, and Zed’s collaborative editor, each aiming to build AI-native infrastructure rather than incremental GitHub alternatives.
Cursor’s Origin: Git-Compatible Hosting for Agent-Heavy Workloads
Cursor’s new project, Origin, is pitched as a Git-compatible AI code hosting platform built for a world where agents do most of the work. Origin emerges from Cursor’s acquisition of code-review startup Graphite, whose customers like Shopify, Snowflake, Notion, and Figma were already stressing traditional workflows with AI tools. Co-founder Tomas Reimers argues that AI has “enabled every developer to be a 10 to 100x developer,” but that this output demands “fundamentally different tooling.” Origin, still in waitlist mode, is described by commentator Shawn Wang as a “long-awaited Git competitor, scalable for agent workloads, extensible with API and MCP, and with built-in merge conflict and CI failure agent resolution.” The timing is strategic: SpaceX has agreed to acquire Cursor’s parent company Anysphere in an all-stock deal valued at USD 60 billion (approx. RM276 billion), as it chases vertical integration of compute infrastructure, models, and applications across its AI stack.
GitLab’s AI-First Infrastructure: Incremental Tool Becomes Platform
While Cursor rebuilds hosting from scratch, GitLab is pushing its existing DevSecOps platform toward AI-first infrastructure. The company already sits in many enterprises as a central code repository platform, which gives it a path to evolve into one of the most credible GitHub alternatives for AI workflows. Instead of replacing Git outright, GitLab layers AI-assisted reviews, automated pipelines, and policy enforcement on top of familiar concepts like branches and merge requests. That makes it attractive to teams that want AI agent infrastructure without abandoning established compliance and governance patterns. In this model, AI agents become first-class CI/CD participants and security scanners rather than external bots spraying pull requests. GitLab’s bet is that enterprises will prefer gradual migration: keep current repos and workflows, but shift the operational center of gravity from human-driven merge queues to AI-orchestrated pipelines that can handle the rising volume of machine-written code.
Zed and the Agent-Centric Editor Stack
Zed approaches AI code hosting from the opposite direction: start with the editor, then reshape collaboration around agents. Designed as a fast, multiplayer code editor, Zed already treats real-time human collaboration as a baseline capability. As agents enter the loop, that collaborative core becomes a way to host AI within the development session itself. Instead of thinking in terms of pull requests and branches, Zed encourages continuous, shared workspaces where agents can propose changes, run tests, and comment alongside humans in near real time. Code hosting becomes an implementation detail behind a live environment. For AI code hosting, this matters: AI agents need low-latency context and bidirectional communication more than static diffs. Zed’s architecture hints at repositories that look less like archives and more like synchronized state logs, where agents and humans co-edit, and “merging” is replaced by a stream of applied, reversible operations.
A Post-GitHub Architecture: Repos for Machines, Not People
The competition between Cursor’s Origin, GitLab, and Zed signals a deeper shift: code repositories are being re-architected for machines first, humans second. GitHub’s infrastructure, which helped start the AI coding era with Copilot, is now “buckling under the weight of it” as incidents pile up and developers move reviews into tools like Claude or Codex. Origin’s focus on agent-scale Git, GitLab’s AI-infused pipelines, and Zed’s live multiplayer editor present three competing blueprints for AI code hosting. All treat AI agents as always-on collaborators that need APIs, context, and authority, not as external plugins. For teams, this reshapes GitHub alternatives from niche choices into strategic decisions about how much of their software lifecycle they want AI to run. The emerging AI agent infrastructure will likely define not only where code lives, but who—or what—has the final say on changing it.
