GitHub recorded 7.38 billion commits in a single month — September — more than five times the previous year’s total. That surge did not come from human developers alone. AI coding agents are pushing commits around the clock, and the platform’s storage layer is buckling under the load.
GitHub’s answer: retire Spokes, the system that has replicated every repository for years, and rebuild Git storage from scratch. Early internal tests of the new architecture point to a 35x write performance improvement, according to The Register’s coverage of the announcement.
TL;DR: GitHub is replacing Spokes, the disk-based replication system that has backed every repository for years, with a rewritten Git storage architecture designed for AI-agent traffic. The platform logged 7.38 billion commits in September — over 5x the previous year — and early internal tests show a 35x write performance improvement, per The Register.
Why Did AI Agents Break Git’s Scaling Math?
AI coding agents are generating commit volume at a scale GitHub’s infrastructure was never designed to absorb. In September alone, the platform recorded 7.38 billion commits — more than five times the previous year’s total, according to DevOps.com. That is not gradual growth. That is a step change.
The agents behave differently from human developers. A single AI agent can push commit after commit in rapid succession while iterating on code, and coverage of the announcement describes this as an unprecedented pattern of demand hitting Git storage directly.
Git was built around human workflows: a developer writes code, commits, maybe pushes a few times per hour. Agents compress that loop into seconds. Multiply that behavior across millions of repositories, and the storage backend faces write volumes that the old architecture simply was not sized for.
The result is the situation sources describe: infrastructure built for a different era of traffic now responding to a surge driven by agentic workflows. GitHub’s rebuild is a direct answer to that shift.
What Is Spokes and Why Is GitHub Retiring It?
Spokes is the architecture that, for years, replicated every GitHub repository across disk-based storage replicas. Coverage from El Ecosistema Startup describes GitHub as now substituting that architecture — the layer that sustained Git by keeping repository copies synchronized on disks.
The system served the platform through years of growth. But the arrival of agentic traffic has exposed its limits. The Register’s reporting frames the rewrite as a move to restore reliability just as demand from AI agents reaches levels the company calls unprecedented.
Why retire rather than patch? Sources describe GitHub as rebuilding its Git storage architecture from scratch — a full rewrite, not an incremental upgrade. That choice signals the scale of the problem: the traffic patterns agents produce are different enough that adapting the old disk-replication model was apparently not the path taken.
For users, the practical stakes are reliability. Every push, clone, and fetch depends on the storage layer staying consistent across replicas. As commit volume multiplies fivefold year over year, the cost of keeping that consistency grows with it — and GitHub has decided the foundation itself needs replacing.
How Much Faster Is the New Storage Architecture?
Early internal tests of the new architecture show a 35x write performance improvement, according to The Register and DiarioBitcoin’s coverage. That figure applies to writes specifically — the operation agents hammer hardest.
Write-heavy workloads are exactly what agentic coding produces. An agent iterating on a repository generates a stream of pushes that dwarfs what a human-driven repo sees. A 35x improvement on that axis is the headline number GitHub is sharing from its early testing.
Some context matters here. These are early internal test results, not production benchmarks on live traffic. Sources consistently describe the 35x figure as coming from those first tests with the new architecture, and GitHub’s rollout details remain subject to what the company discloses next.
Still, the direction is clear. The rewrite is explicitly aimed at handling demand that sources describe as unprecedented. If the early numbers hold at scale, the storage layer would finally be sized for the traffic pattern agents actually generate — rather than the one human developers produced.
What Changed About Traffic Patterns on GitHub?
The headline change is raw volume: 7.38 billion commits in September, over five times the previous year’s total. Coverage from Digital Today describes GitHub as overhauling its repository architecture specifically after this surge in AI agent traffic.
But volume is only part of the story. Sources emphasize that the demand comes from AI agents — automated systems that operate continuously and independently of human working hours. When agents push commits at machine pace, the traffic profile stops looking like human developer activity and starts looking like something closer to service workload.
That shift has consequences for how storage must behave. Replication, consistency checks, and failover all become harder as write rates climb. The Register describes GitHub as responding to this unprecedented demand with a from-scratch rebuild of Git storage — a reaction to the pattern itself, not just its size.
The question the industry is now watching: as agent adoption grows across platforms, how many other infrastructure layers built for human-paced workflows will need the same treatment? GitHub’s rewrite is the first major answer.
Frequently Asked Questions
How many commits did GitHub record in September?
GitHub recorded 7.38 billion commits in September, according to DevOps.com. That figure is more than five times the previous year’s total, a jump attributed largely to traffic from AI coding agents.
What is Spokes?
Spokes is the architecture that for years replicated every GitHub repository across disk-based storage replicas, as described in El Ecosistema Startup’s coverage. GitHub is now replacing it with a rewritten Git storage architecture.
How much faster is GitHub’s new storage architecture?
Early internal tests show a 35x write performance improvement, according to The Register. The figure comes from GitHub’s early testing with the new architecture, which the company says responds to unprecedented demand from AI agents.
Why are AI agents stressing GitHub’s infrastructure?
AI coding agents push commits at machine pace, iterating on code continuously rather than a few times per day. The result, per coverage of the announcement, is commit volume more than 5x the previous year — traffic GitHub describes as unprecedented and its storage architecture as unprepared for.
Why Does Storage Reliability Matter More Now?
Because the volume of writes hitting GitHub has exploded to a level the old architecture was never designed for. GitHub recorded 7.38 billion commits in September 2026, more than five times the previous year’s total, and sources attribute that surge directly to AI coding agents pushing commits continuously. Each agent session can generate commit after commit, turning what used to be occasional human-driven writes into a constant, machine-driven stream.
Reliability is the casualty here. Spokes, the system that for years replicated every repository across disks, was built around assumptions about write frequency that no longer hold. When millions of automated agents work in parallel, even small per-write inefficiencies compound into infrastructure-level strain.
The stakes are higher than they sound. GitHub is the world’s largest host of source code, so storage degradation doesn’t stay contained — it ripples into CI pipelines, code review flows, and every downstream dependency. The rewrite is a bet that storage must be redesigned for machine-scale writes, not merely patched.
Is fivefold growth in a single year sustainable on decade-old assumptions? GitHub clearly doesn’t think so.
What Does the Rewrite Mean for Developers?
For developers, the most visible change is speed. Early tests with the new architecture show a 35x improvement in write performance, a figure GitHub itself has highlighted as “Octocats claim 35x write improvement” in its public messaging. That means faster pushes, quicker agent commits, and less waiting on infrastructure during peak load.
For teams running automated workflows, the difference could be substantial. An agent that commits dozens of times per session multiplies even small write latencies into noticeable friction. Reducing write overhead attacks that friction at its source.
What developers should not expect is an immediate, all-at-once cutover. Sources describe internal testing and early validation phases, which suggests the rollout will be gradual rather than a day-one replacement. Repositories keep working as they always have — Git’s protocol doesn’t change.
In my opinion, the deeper benefit is confidence: if the storage layer is rebuilt around actual demand rather than historical assumptions, the reliability problems that motivated the rewrite should fade over time rather than resurface under load.
Is This a Sign of Broader Infrastructure Strain From AI Agents?
Yes — the evidence points that way. The title of the underlying coverage says it plainly: “Coding Agents Broke Git’s Scaling Math.” That is not a metaphor dropped casually; it reflects GitHub’s own acknowledgment that AI agent traffic forced an unprecedented, from-scratch rebuild of its Git storage architecture.
Consider what the numbers imply. 7.38 billion commits in one month, more than five times the prior year, driven substantially by autonomous agents. Human developers didn’t suddenly quintuple their output overnight. Machines did.
This pattern is likely to generalize. Any platform whose write volumes were modeled on human pacing now faces workloads where software commits around the clock. Sources say GitHub is responding to “unprecedented demand from AI agents,” a phrase that suggests the company itself views this as uncharted territory.
The lesson for the broader industry is uncomfortable: agent-generated load doesn’t arrive gradually. It arrives as a step function, and infrastructure built on historical growth curves gets caught flat-footed.
What Happens to the Old Spokes Architecture?
Spokes is being replaced. Sources describe GitHub “substituting” the architecture that for years replicated every repository across disks, and The Register’s headline is blunt about it: “So long, Spokes.”
Spokes served its purpose for a long time. It replicated repositories to multiple locations to protect data and keep the service available during failures. But the model of duplicating every repository onto physical disks across replicas scales poorly when write volume grows more than fivefold in a single year.
The transition appears staged rather than abrupt. Coverage describes early internal tests showing the 35x write improvement, which means the new system is being validated before taking over production traffic. That is the standard playbook for replacing a load-bearing component: prove the successor first, then retire the predecessor.
What happens to the name itself is a symbolic matter. The more meaningful question is when Spokes fully exits service — and on that timeline, sources only indicate the rewrite is underway, not a specific completion date.
Can Other Git Platforms Learn From This?
Yes, and the central lesson is architectural: storage layers designed for human-scale writes will break under machine-scale writes. GitHub’s decision to rebuild “from scratch” rather than optimize incrementally signals that patching was not seen as sufficient once agent traffic changed the growth curve.
Other platforms can draw several practical takeaways:
- Model capacity on agent write patterns, not historical human commit rates
- Treat a fivefold year-over-year growth spike as a redesign trigger, not a tuning exercise
- Measure write amplification across replication layers under automated workloads
- Invest in write-path performance early; GitHub’s early tests show 35x gains are achievable
- Plan staged migrations so reliability improvements don’t require risky cutovers
- Assume autonomous agents will dominate commit volume, not just contribute to it
- Build observability that distinguishes agent traffic from human traffic
- Revisit the assumption that replication across disks is the right reliability primitive
Whether smaller platforms face the same magnitude of load is another question. But the direction of the industry is clear: agents commit more, and commit more often, than the teams that deploy them.
Frequently Asked Questions
How many commits did GitHub record in September 2026?
GitHub recorded 7.38 billion commits in September 2026. Sources report this was more than five times the previous year’s total, with the surge attributed largely to AI coding agents pushing commits continuously.
What is Spokes?
Spokes is GitHub’s long-standing storage architecture that replicated every repository across disks to maintain data durability and availability. GitHub is now replacing it with a storage layer rebuilt from scratch in response to unprecedented AI agent demand.
How much faster is GitHub’s new storage architecture?
Early internal tests show a 35x improvement in write performance compared to the current system. GitHub has publicly highlighted this figure, describing it as a 35x write improvement achieved with the new architecture.
Why is GitHub rewriting its storage now?
Because AI coding agents broke the existing scaling math. With 7.38 billion commits recorded in September alone — over five times the previous year — the Spokes architecture’s disk-replication model was no longer adequate, prompting a ground-up rebuild to restore reliability.
Summary
GitHub’s storage rewrite is one of the clearest signals yet of how AI agents are reshaping core infrastructure:
- Commit volume has exploded: 7.38 billion commits in September 2026, more than five times the previous year’s total, driven substantially by AI coding agents.
- Spokes is being retired: GitHub is replacing the architecture that replicated every repository across disks, described bluntly in coverage as “So long, Spokes.”
- Writes are dramatically faster: Early internal tests show a 35x improvement in write performance in the new architecture.
- The rebuild is from scratch: Rather than incremental optimization, GitHub is reconstructing its Git storage layer to handle machine-scale, around-the-clock write traffic.
If you run automated workflows or agent pipelines against GitHub, watch the rollout closely — and start modeling your own infrastructure on agent-paced writes, not human ones. The scaling math has changed for everyone.
Sources: DevOps.com, The Register, Digital Today