From Pablo Santos Luaces · · 3 min
Version Control's Second Explosion
AI coding agents and giant monorepos are triggering the biggest shift in version control since Git launched in 2005.
In brief
AI coding agents and giant monorepos are triggering the biggest shift in version control since Git launched in 2005. Version control is moving beyond local Git clones toward cloud-first, agentic tools built for high-frequency commits and giant monorepos. Originally reported by Pablo Santos Luaces.
A long quiet period ends
In 2005, Linus Torvalds created Git, launching two decades of platform stability. Now, a fresh wave of innovation is disrupting code management once again.
“I hadn’t felt that same electricity again in twenty years. Until a few months ago.”
AI agents drive the shift
Automated tools write code and issue commits at super human speeds. Systems built for manual developer workflows struggle under this massive commit traffic.
“Two big pushes happening simultaneously: the need to handle a much higher commit pace due to agents, and a sudden urge to replace GitHub.”
Market leaders face disruption
GitHub shaped software collaboration for over fifteen years. Yet platforms like SourceForge once seemed unassailable before fast shifts rendered them obsolete.
“GitHub is still gigantic in both repos and minds, and its impact in the world of software development and collaboration in general has been enormous”
Lessons from past tool wars
Legacy tools like ClearCase offered virtual file systems and powerful branching, but steep costs killed them. Git filled the void with affordable, fast local branches.
“Clearcase was extremely expensive, my memory tells me something around 5k/year per user for an enterprise setup. The version control bug bit me hard.”
The 2005 explosion set standards
Replacing Bitkeeper sparked a race between Git, Mercurial, and commercial tools. GitHub eventually won the market by wrapping Git in a social web platform.
“It was a revolution, and a little bit later, around 2008, GitHub brought it to the masses and the rest is history.”
Entire tracks agent provenance
Founded by GitHub's former CEO, Entire records which AI agent and prompt created every line of code while speeding up worldwide repository cloning.
“They have released provenance tracking to basically know what code which agent created and which prompt”
Pierre scales repository storage
Pierre built code.storage to keep up with intense traffic from AI code generators, targeting the bottlenecks GitHub hits under high commit volume.
“GitHub hit a wall in terms of performance when dealing with an incredibly high increase of commits, PRs, and traffic in general”
Origin speeds up Git hosting
Built by the team behind Cursor, Origin redesigns Git hosting architecture around write-ahead logs and cloud storage to push speed to absolute limits.
“This time, instead of racing against Git, I’m trying to make it as fast as I can.”
GitButler updates the desktop UI
GitButler reimagines the client interface with virtual branches and stacked commits, tailoring the local workspace for AI assisted workflows.
“Stacked branches (we all need them in my opinion), better rebasing and as I mentioned, innovative ideas”
Jujutsu standardizes tracked conflicts
Google created Jujutsu as a successor to Git. It introduces native stacked branches and tracks unresolved merge conflicts directly inside the repository.
“It brings innovative ideas like stacked branches, and goes beyond that versioning even unsolved conflicts in the repo”
Diversion drops local clones
Diversion skips local cloning entirely in favor of cloud-first workspaces, serving multi-terabyte game repositories with instant server sync.
“At that size, having local clones is simply a no go. Even more, the fact of ‘commit then push’ was a drag in many professional environments”
Oxen manages dataset versions
Oxen focuses on versioning massive AI datasets and raw media. It relies on low level optimizations to transfer large files without friction.
“Oxen is not in the same forge race as the others, but they are a good example of the fact that the market is evolving more than it did”
Epic Games releases Lore
Epic Games created Lore, an open source system built without Git internals. It streams game monorepos and patches only changed binary data.
“It is a complete full stack, not based on Git, which makes it very different and exciting.”
Distributed models hit scale limits
Cloning full history locally breaks down when monorepos grow beyond hundreds of gigabytes. Large codebases force teams back toward server-centric sync.
“My bet is that distributed with push/pull instead of simply commit will be limited by repo sizes”
Main branches need parallel writes
Modern Git hosts serialize incoming commits, creating severe bottlenecks. Next generation engines are revisiting parallel write models for high volume repos.
“Right now committing to main on Git forges requires some serialization, so there is a limit in the speed of those commits.”
Virtual file systems return
Hydrating files on demand lets engineers navigate giant codebases using minimal disk space, though operating system limits still pose technical hurdles.
“There are some barriers for this to happen, though: first is macOS, which makes it virtually impossible to develop fast virtual file systems”
The era of coping ends
For twenty years, developers simply adapted to Git constraints. High volume AI commits and massive monorepos are finally opening up new possibilities.
“For twenty years the interesting question in version control was “how do we cope with Git?” Now it is “what comes after?””
The essence
Version control is moving beyond local Git clones toward cloud-first, agentic tools built for high-frequency commits and giant monorepos.
