AIToday
Large Language ModelsAI Coding AssistantsAI Business & IndustryPublickeyPublished: Nov 4, 2026, 01:00 JST

GitHub: 180x throughput in AI agent test, all users by early next year

Send on LINE
GitHub: 180x throughput in AI agent test, all users by early next year

3 Key Points

  1. What happened

    At GitHub Universe 2026, Microsoft CoreAI EVP Jay Parikh said a test with 5,000 AI agents in one repository handled 7,200 writes per second, 180 times the old architecture.

  2. Why it matters

    The jump comes from locking per branch instead of per repository and splitting compute from storage, so AI agents on different branches can work in parallel.

WHO IT HITSDevelopment teams running AI coding agents on shared repositories should see far fewer write bottlenecks, and platform engineers planning large agent fleets can expect branch-level concurrency to matter more than repository-level limits.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

GitHub has been expanding its infrastructure at a rapid clip, adding 4 million CPU cores, 160 PB of storage, and 50 Tb/s of network bandwidth this year alone, while also migrating onto Microsoft Azure with a multi-region rollout that is set to finish early next year.

The architectural change described at GitHub Universe 2026 goes further than raw capacity. By keeping code in Azure Blog storage and letting Git servers act mainly as fast caches, GitHub says it can scale storage and servers independently and keep coordination between them to a minimum. Only a small number of reference transactions remain on the critical path, handled by a dedicated log service.

The practical effect shows up in the write path. Under the old architecture, pushes to a single repository topped out at a few dozen per second; the new design targets thousands per second. One likely implication is that AI agents become a first-class workload for GitHub rather than an add-on, since branch-level locking lets agents on separate branches work concurrently — a shift that may reshape how teams organize repositories as agent fleets grow.

FAQ
What specifically did GitHub change to get this speedup?
GitHub separated compute from storage so servers only cache data, and it moved write locks from the repository level to the branch level.
When will regular GitHub users get the new architecture?
GitHub plans to migrate all users to the new architecture by early next year.
What did the 5,000-agent test measure?
With 5,000 AI agents working in a single repository, the system processed 7,200 writes per second — 180 times the previous throughput.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleBCG: AI build-out needs $2.6 trillion a year by 2030