GitHub rebuilt core parts of its Git infrastructure to handle a surge in traffic from AI coding agents working alongside human developers, according to a GitHub Blog post by principal software engineer Brian Celenza. Monthly commits grew from 1.3 billion to 7.38 billion in a year, and monthly pushes grew nearly fivefold, from 690 million to 3.35 billion, Celenza said.
GitHub's busiest single repository took in about 1 billion requests in August, and total Git activity across the platform roughly doubled, from 218.2 billion to 473.3 billion events a month, Celenza wrote. GitHub Actions ran 3.26 billion times in September, about four times its volume a year earlier, according to the post.
"Writes are way harder," Celenza wrote. "Every push has to be stored durably and made visible consistently before the next agent or CI job can build on it." He said GitHub redesigned its backend around two principles: minimizing coordination by moving maintenance work off the serving path, and decoupling storage from compute by moving authoritative repository data to Azure Blob Storage while lightweight caching workers handle reads.
Celenza said internal benchmarks showed up to 35 times higher write throughput under the new design, with read capacity that scales independently of writes.
Git was built assuming a human commits every few minutes, not a fleet of agents pushing continuously. GitHub rebuilding its storage layer around that assumption is a quiet admission that agentic coding has already changed the traffic pattern, not just the tooling, for anyone running a busy repository.