--- title: "AI Generates So Much Code That Anthropic Can't Keep Up — Forced to Invent Its Own Git" description: "Anthropic's Claude models generate so much code that the company created its own version control system, Scion, to prevent infrastructure collapse." date: 2026-10-15T13:02:23.465Z lang: en url: https://xab.info/en/posts/ai-generates-so-much-code-anthropic-invented-git tags: [anthropic, claude, artificial-intelligence, git, version-control, programming, software-development] author: "Alexander Buran, Roman V." publisher: "XAB.info" --- # AI Generates So Much Code That Anthropic Can't Keep Up — Forced to Invent Its Own Git ![Development and code management interface at Anthropic](https://xab.info/media/2026/10/15/ii-generiruet-stolko-koda-chto-anthropic-sozdala-svoy-git/ii-generiruet-stolko-koda-chto-anthropic-sozdala-svoy-git-1.webp) ## Key Points - Code merges at Anthropic surged from 2,500 to 30,000 per day thanks to AI. - The company developed its own API-based version control system called Scion. - The Claude Mythos Preview AI model helped optimize the data migration pipeline. - A public open-source release of the Scion system is not planned. Advanced artificial intelligence models of the Claude family, developed by Anthropic, have reached an unprecedented level of autonomy and productivity. The rate of software code generation has become so high that standard development infrastructure could no longer cope with the avalanche of data. While internal systems recorded about 2,500 code merges per day in May of this year, this figure skyrocketed to a colossal 30,000 daily merges by September. As a result, the company's primary bottleneck shifted from writing code to managing it. ### Creation of the Custom Scion System To overcome this infrastructural crisis, the company's engineers were forced to develop their own specialized version control system called Scion, operating on a single giant monorepo. The migration process for all workflows took exactly five months. Scion's architecture was originally designed with AI agents in mind, shifting the lion's share of computational load from client machines to the server side. Interactions with the environment are now handled via API, eliminating the need to store local repositories and drastically reducing network traffic volumes. ### AI Assisting in Infrastructure Software Development Interestingly, artificial intelligence and advanced company models played a direct role in developing Scion itself. When the advanced Claude Mythos Preview tool became available internally, engineers tasked the neural network with optimizing the data migration pipeline from classical Git to the new Scion format. The result exceeded expectations: the AI-created algorithm consumed twenty times less RAM and operated four times faster than the human-written prototype. ### The Scale of Infrastructural Challenges Traditional source code management approaches, including the decentralized Git model, proved ineffective at Anthropic's scale. The main branch of the corporate repository has grown to about 100 gigabytes, with a compressed working copy reaching a quarter of that size. The infrastructure processes over 25 million code retrieval operations daily. If standard network synchronization methods were applied, Git's daily inbound traffic would inevitably exceed 600 petabytes, paralyzing the company's entire IT ecosystem. ### Controversial Data It is worth noting certain debatable aspects surrounding AI autonomy metrics in the industry. While Anthropic openly reports a radical surge in machine-generated code and solves internal infrastructural problems by creating unique utilities like Scion and Antique, independent experts express concerns over the risks of diminishing deep human control over the codebase. Anthropic engineers themselves emphasize that they are comfortable with AI outperforming humans in coding, yet they maintain strict oversight over neural network architectural initiatives, cutting off any attempts by autonomous systems to violate safety parameters and integration protocols. ### Prospects and Company Plans Anthropic's leadership has officially confirmed that it does not plan to release the Scion system as an open-source project. Nevertheless, given the rapid development of generative artificial intelligence, other technological giants will face similar infrastructural challenges in the near future. By sharing its unique experience, the company aims to push the industry toward rethinking version control approaches in the era of neural network dominance. ## Fact-Check & Sources - [ИИ генерирует столько кода, что Anthropic не успевает его обрабатывать — пришлось изобретать собственный Git window-new](https://3dnews.ru/1149906/ii-generiruet-stolko-koda-chto-anthropic-ne-uspevaet-ego-obrabativat-prishlos-izobretat-sobstvenniy-git) - Основной источник о проблемах инфраструктуры Anthropic и создании системы Scion. - [Anthropic заявила, что ИИ уже пишет более 90 % её кода](https://3dnews.ru/1141738/anthropic-zayavila-chto-ii-uge-pishet-bolee-90-eyo-koda) - Основной источник о проблемах инфраструктуры Anthropic и создании системы Scion. ## FAQ ### Why did Anthropic create its own Git? Existing version control systems could not handle the massive volume of code generated by Claude models. ### What is the Scion system? It is a specialized monorepo-based version control system working via API and optimized for AI agent workflows.