Anthropic builds Scion, a git replacement, as Claude pushes merges to 30K per day
The AI company replaced git with custom version control as autonomous code generation created a 12x surge in daily merges.
Part of a larger narrative
Anthropic's Claude Evolution
- Anthropic builds Scion, a git replacement, as Claude pushes merges to 30K per dayyou are here
- Anthropic warns small AI models may soon self-replicate on smart devices
- Anthropic cuts Claude reasoning-token theft by up to 99%, flags Chinese rivals
- Anthropic expands free Claude for Startups program with $1,000 in credits
- AI Whistleblower Coxon to Testify Before NYC Council Alongside Tech Giants
What to know
- Anthropic's autonomous code generation scaled merge volume 12-fold in four months (2.5K to 30K+ daily), breaking git's design assumptions and forcing a ground-up rewrite of version control infrastructure.
- Claude itself built Scion's most critical component—rewriting the ingestion pipeline with 20x better memory efficiency and 4x speed with minimal prompting—turning the AI bottleneck into the solution.
- Anthropic frames this as a preview: other heavy users of AI coding tools will soon face identical infrastructure strain, suggesting a wave of custom VCS implementations across the industry.
- The post raises a philosophical question about human agency in software development: with agents autonomously generating and merging tens of thousands of changes daily, what remains for human engineers to do?
The dispute Whether Anthropic's 30K+ daily merges represent genuine software engineering progress solving real problems, or a tautological loop where removing bottlenecks merely increases code churn without output. · positions read across 147 posts and comments
AI's capability to scale code generation and self-optimize infrastructure is profound and represents a genuine inflection point in development practice.
-
“I feel like the advantage of having these models in house can't be understated. Companies with access to inferior models, tr…”
trnvll · Hacker News ↗
The claimed infinite merge demand and code velocity don't align with solving real-world problems; something about this scale is philosophically incoherent.
-
“If the goal is to write software which solves real problems, how can the demand for code grow infinitely?”
netburst · Hacker News ↗
Other companies will rush to build similar systems and suffer the maintenance burden, not realizing they lack Anthropic's scale or internal AI models.
-
“There's going to be an ugly phase where everybody FOMOs into thinking they need to do the same, only to realize that they're not Anthropic/Google/etc.”
preommr · Hacker News ↗
How it unfolded 1 development · click the chart to see its coverage articlespostscomments
-
1
Anthropic publishes technical details of Scion architecture
Anthropic released a blog post describing Scion's design and why git—which stores 100 GB of history and 25 GB of compressed code in their monorepo—became impractical. The post notes Claude rewrote the ingestion pipeline to be 20x more memory-efficient and 4x faster with no further instruction.
“Claude's ability to write code was both what put us in this situation in the first place and how we were able to quickly build a high-quality, scalable system to deal with it.”
— Anthropic -
6 outlets Redesigning source control at Anthropic (via) Interesting piece from Anthropic describing …
first by Simon Willison's Weblog, 1d ago · also The Register, Slashdot, Simon Willison, HN Best, HN Frontpage
2 more headlines
- Anthropic reinvents source control as Claude floods the world with code The Register · 1d ago
- Redesigning source control at Anthropic Simon Willison · 1d ago
-
Version control software is something that you _really_ want to be reliable. Anthropic has a released a lot of vibeslop, and a lot of it is pretty bad. Claude Code has an endless series of terminal-drawing bugs. And even when they're fixed, they come back a week later. This is not the sort of thing that I want in my version control system. Git…
2 more of the top 3 · 147 posts in this stretch
-
C
Anthropic, of course, is not just riding down the vibe bobsled, they have rocket packs attached to the bobsled, careening at record speed https://www. anthropic.com/engineering/rede signing-source-control
-
great article. Besides the source control stuff, I found this very interesting (and matches my experience with agentic coding in general, now we'll see until hiring pipelines will take this kind of skillset into account). I am not sure however how easy it is to review 30k PRs/day even if only for intent, and how one can easily notice if the model…
-
-
background
Anthropic completes migration to Scion — In five months, a team of seven engineers with Claude's assistance completed a full migration of Anthropic's monorepo from git to Scion, also shipping a code review system (Antique), code search, and IAM tooling.
-
background
Merge rate surges to over 30,000 per day — By September 2026, Anthropic's daily merge count grew to over 30,000—a more than tenfold increase since May—as Claude autonomously iterates on research loops and builds services at scale.
-
background
Anthropic begins building Scion replacement for git — Facing exponential growth in merges, Anthropic starts building an in-house version control system called Scion, designed for centralized, high-throughput operation.
-
background
Anthropic's merge rate hits 2,500 per day — By May 2026, Anthropic's monorepo averaged around 2,500 code changes merged per day, already straining git's design.
What people are saying 21 voices from 5 sites · best of 147 · verbatim
- What problems are being solved by the 30,000+ daily code changes, and how does solving problems lead to infinite merge demand?
- Will other companies that build similar VCS systems face similar total-cost-of-ownership challenges, and do they have access to Claude-tier models to maintain the infrastructure?
- Yesterday
-
I came into this blog post a hater, but honestly - hats off to the team. This sounds like it was a very arduous project, and I don't envy the engineers that have to sweat over handling Anthropic's scaling commit volume.To that end though, this line feels like it (intentionally?) undersells the engineering team -> Very often, we found that the role…
-
A
Anthropic posted about this yesterday: Anthropic: Redesigning source control at Anthropic https://www. anthropic.com/engineering/rede signing-source-control "Anthropic's Claude models have been emitting so much code that the company's coding infrastructure hasn't been able to keep up." "It's a symptom of the AI coding imperative and…
-
I'm not so sure it works even if you have a billion dollars. Claude Code is a bug riddled mess - e.g. on my Mac, I've never been able to copy more than one line of output. This is arguably fine for Claude Code since clearly people are willing to tolerate it (although I think they would improve the company value quite a bit by focusing on fixing…
-
D
OpenAI's decision to build its own SCM tool is not the flex it thinks it is. In fact, it reveals a development process that has gone off the rails. 30K commits per day is impossible for humans to review, so clearly they are not even bothering. They should dispense with "source" control for Claude output and treat it as object code. If AI is such…
-
Like one proverb says: One man's shit is another man golden standard. Not everyone has the same standard but most people would say it's a shit architect. A confident shit architect for sure.
-
It’s the difference between the product and the workflow. “Good enough for now” has always been part of a prudent development plan. You don’t know what the user base will be, what demand will be, etc, so there’s no sense in building a golden palace to support those assumptions. The tooling, workflow, review, test, incident management - you don’t…
-
This is exactly the right skepticism. Merges per day is a throughput number - it tells you how fast code lands, not whether it was good. The number that would convince me is the revert rate, or how often CI catches something after merge. I've felt this in miniature on my own projects: the moment generation gets faster than my ability to actually…
-
Can you dm or tell me how you use two 20x plans in parallel? Just have one plus codex and it’s fine but looking to step it up
-
> one of the most annoying habits of unsupervised LLMs The key word is _unsupervised_. If a person writes down the detailed requirements, and ensures that the LLMs follow a good design, it usually ends up with only 60,000 lines of buggy Go full of 3 subtly different versions of common functions. :(
-
That works if you have billions of dollars. The average SME likely doesn't have enough people to "AI prompt" all of the software that they need. The ongoing maintenance has a non-zero personal cost, which is likely larger than just buying the software.
-
Most of the world isn't perfect, and trying to be perfect means you get nowhere. In most places, excessive perfection is implicitly avoided through resource constraints.In the end, what matters is whether something fulfills its purpose, not how it does it. Code quality, understanding code, etc. are just a means to this goal.Some level of poor…
-
I think at work I am just barely in the category of I've removed more code than I have added. Over many, many bugs and long meetings I've sort of arrived at the conclusion that in most cases, the best code (excluding comments) is the code that doesn't have to exist [0]. It occurred to me at work today while refactoring some change that even in…
-
I can't rationalize why they write these long, technical blog posts without sharing anything of what they've written. What are they afraid of? Meanwhile, I can't believe I'm saying this, but I have to give Meta credit for making a similar post, but not being afraid to share it openly:
- Oct 13
-
A company with (I'm guessing) less than 2500 developers need to make tens of thousands of merges every day? And they somehow had a team of 7 Claude baby sitters handle this entire thing? I use Claude everyday and I'm predicting this will turn into a major clusterfuck in not too long. Even taking into account the fact that they can burn virtually…
-
I would’ve loved to see the landscape review they did of other VCSes before building their own. Cynically, yes of course this is content marketing to show off for investors and customers. But even taking it as content marketing I’d expect to see a comparison with other solutions. Let alone taking it at face value as a technical blog post, which…
-
At least they admit they're the reason Github is breaking. 30.000 LLM-sized commits to a single repo every day with agents hammering git blame as well, my god. I don't want to know what kind of useless onslaught the Github infrastructure has to withstand in total. https://www. anthropic.com/engineering/rede signing-source-control Of course they're…
-
P
Any startup that hires enough ex-Googlers ends up with an ad hoc, informally-specified, bug-ridden, slow implementation of half of Google infra https://www. anthropic.com/engineering/rede signing-source-control
-
C
Anthropic now states that Claude can code better than its engineers but humans are still valuable primarily to tell Claude “no.” This was shared as part of their post about a small team building a replacement for git in five months. https://www. anthropic.com/engineering/rede signing-source-control
-
Many parts of the Git distributed version control are the results of good decision made. Anthropic source control team just think that their trade-offs needs are vastly different from other businesses. They said these too:1. they have no-cap token spend allowance internally2. their agent swarms are amongst the largest if not the largest, again…
-
This is pretty insane. Anthropic was committing so much Claudeslop (2,500 changes a day) that it literally broke their Git Monorepo. They had Claude write a new version control system called Scion where all the commits happen on the server, since the AI agents don't work offline anyway.
-
Anthropic's code changes merged per day has 10x since May. 'Claude is able to autonomously iterate on research loops, micro-optimize complex programs, and build new services from the ground up at a speed and scale we haven't seen before.'