---
title: 'In the News: November 7, 2026 (Midday)'
description: 'GitHub says its new Git storage design handled 5,000 concurrent agents sending 7,200 pushes per second to one repository in internal tests.'
canonical_url: 'https://darkfactory.dev/news/2026-11-07-midday'
markdown_url: 'https://darkfactory.dev/news/2026-11-07-midday.md'
collection: news
date_published: '2026-11-07T13:10:00-05:00'
date_modified: '2026-11-07T13:10:00-05:00'
---

# In the News: November 7, 2026 (Midday)


One item at midday: GitHub's engineers explain how they rebuilt Git storage so that many agents can write to one repository at the same time, with the numbers from their own testing.

## 1. GitHub rebuilt Git storage for thousands of agents pushing to one repository

**[Inside GitHub's new Git storage architecture](https://github.blog/engineering/architecture-optimization/inside-githubs-new-git-storage-architecture/)** · Karthik Raman and Brian Celenza, GitHub · GitHub Blog, November 5, 2026

GitHub reports that internal testing of the new design, code-named Rubicon, supported 5,000 concurrent agents working on one repository and 7,200 pushes per second to it, with consistently low read latency. The authors say the busiest repositories on GitHub.com today receive at most dozens of writes in a single second, in short bursts, and that they expect agent-driven work to turn those bursts into a sustained rate within the next year. Each push also triggers reads from CI, deployment systems and other agents, so read traffic grows with writes.

The design rests on one observation: Git objects are immutable and named by their contents, so only the reference update needs agreement between writers. Packs live in Azure Blob Storage. References and the list of packs live in a separate replicated log service, Ref-CAS, which commits each reference update as a compare-and-swap by quorum and locks only the references it touches. The authors say agents mostly push to their own branches, so most updates do not contend. They describe two earlier prototypes that kept references in object storage: one reached write throughput in the mid-hundreds per second, and the other made reads slower because a read had to check every segment. Compaction now runs on separate nodes rather than on the replicas serving traffic. GitHub says Ref-CAS handles reference state alone in the tens of thousands of operations per second.

These are GitHub's figures from internal tests, and the post does not describe the test setup. The architecture is in limited private preview, with a broader rollout planned for early next year. GitHub also lists work it has not shipped: an atomic multi-file commit API, work without a clone, and event streams for agents.

**Why it matters:** GitHub is saying that a single repository becomes a throughput limit once agents write to it continuously, and it expects that within a year. Until the rollout reaches you, the access pattern GitHub designed around is one you can adopt now: each agent pushing to its own branch, so updates rarely contend.