Its next-generation storage architecture supported 5,000 concurrent coding agents pushing to one repository in internal tests. The system processed 7,200 pushes per second while continuing to serve reads with low latency, a workload far beyond the pattern created by human contributors.
The redesign separates immutable Git objects from the much smaller reference updates that must be ordered. Objects are named by their contents, so multiple writers can store them without coordinating. Only changes to references such as branch heads require agreement. That lets storage, request serving and repository maintenance scale independently.
GitHub has also moved virtually all repositories into Azure-backed storage and uses virtual replicas to serve data. For developers, the interface remains ordinary Git: pushes, fetches, branches and repository protections work as before.
The published figures come from GitHub’s own testing rather than an independent benchmark. Still, they illustrate the infrastructure shift caused by software agents, which can create changes and pushes at a rate that repository systems designed around human activity were not built to absorb.