Building Git infrastructure for agent-scale development
GitHub is overhauling its Git infrastructure to support agent-scale development, addressing the growing demands of concurrent agent and developer activity on repositories handling billions of requests monthly.
GitHub is rebuilding its Git infrastructure to accommodate agent-scale software development while the platform remains fully operational. The shift responds to the rising workloads driven by concurrent developer and agent activity in repositories that now process millions of commits daily. Current architecture relies on Spokes, storing full repository copies on local disks with a three-phase commit protocol to ensure consistency. This design has supported billions of repositories but faces limitations at the highest activity levels, where adding replicas to scale reads slows writes.
The busiest GitHub repositories now see roughly a billion requests per month, with total Git activity more than doubling from 218.2 billion to 473.3 billion events between September 2025 and August 2026. In September alone, developers and agents made 7.38 billion commits, over five times the previous year. These top-tier workloads, often involving large engineering teams and extensive CI pipelines, require infrastructure capable of handling sustained, concurrent reads and writes at unprecedented scales.
The existing architecture couples durability and scale, where every push update must be stored durably across multiple replicas before becoming visible. This ensures consistency but creates a tradeoff: adding replicas to absorb read load slows writes, while losing replicas reduces read capacity or halts writes entirely at quorum loss. To meet agent-first demands, GitHub is redesigning the system to separate durability from scale, preserving reliability and governance while enabling independent scaling of reads and writes.
The new architecture aims to deliver higher throughput and reliability by allowing reads and writes to scale independently. Internal benchmarks indicate up to 35 times higher write throughput, with read capacity scaling dynamically to meet demand. The redesign preserves essential controls like branch protections, audit logs, and observability, ensuring teams retain the governance and reliability features they rely on as GitHub evolves to support the most demanding workloads.