Is GitHub Crumbling? How Cursor, GitLab and Zed Are Rebuilding Code Hosting

GitHub’s platform is under severe strain from AI‑generated code, processing roughly 1.4 billion commits and over 17 million pull‑requests each month, far exceeding the 2025 forecast of 1 billion commits. This overload has sparked a wave of rebuilding efforts across the code‑hosting ecosystem.

Origin – unveiled at a private Cursor developer conference in San Francisco – is a Git‑compatible hosting service designed from the ground up for AI agents. Thomas Reimers, co‑founder of Graphite and former head of code‑review startup, announced the acquisition of Graphite in December and said the acquisition accelerates the most ambitious project: rebuilding the tool from scratch for an AI‑first world.

“After being acquired by Cursor, we accelerated our most ambitious project – rebuilding the tool from zero.” – Reimers

Reimers highlighted that Graphite’s customers (Shopify, Snowflake, Notion, Figma) had already felt the pain of AI‑driven workflows before Origin existed. He argued that AI agents have amplified developer productivity 10‑to‑100×, demanding entirely new tooling.

Brian Douglas, former GitHub developer‑advocacy director, now runs AI‑infrastructure startup Paper Compute. He warned that AI agents are “rapidly killing people’s willingness to work on open‑source” and that traditional metrics like commit count no longer reflect value. Douglas proposes using tokens —which map directly to compute cost—as the primary unit for measuring software work.

“Token is a better indicator of value than commit count.” – Douglas

At GitLab’s Transcend conference, chief product and marketing officer Manav Khurana announced Project Switch , a “next‑generation source‑code management” backend that retains the Git protocol but lets AI agents query repositories without full clones. GitLab claims Project Switch can speed up agent tasks up to 50× while reducing token consumption by up to threefold, with Anthropic as a design partner.

“The world’s most popular Git platform is overwhelmed not only by clones and merges but also by dozens, sometimes hundreds, of concurrent agents.” – Khurana

Following GitLab’s announcement, Zed co‑founder Nathan Sobo revealed DeltaDB , a more aggressive redesign that replaces the commit‑based model with a fine‑grained, continuous‑flow architecture where every agent operation is directly tied to the interaction that generated it. A beta is expected within weeks.

SpaceX’s recent $60 billion acquisition of Cursor (valued at $2.5 trillion) adds massive resources to the effort, though it remains unclear whether this will accelerate or hinder the transition.

Cursor has also been developing its own coding model, Composer . Version 2.5, released in May, runs internal inference at a fraction of the cost of Claude Opus, delivering up to ten‑fold fewer output tokens for the same task.

“You can’t just drop an OpenAI key and expect hyper‑growth; you need to own the model to win.” – Douglas

The article concludes that the shift from human‑centric pull‑request workflows to AI‑driven code generation forces a re‑thinking of development infrastructure, IDE roles, and value measurement. Companies that can provide scalable, agent‑friendly hosting and own performant models are positioned to become the next generation of Git competitors.