Nadella: 5,000 Agents Pushed 7,200 Writes a Second, 180x

GitHub pointed 5,000 AI agents at a single code repository in testing and sustained 7,200 writes per second, Microsoft chief executive Satya Nadella said in his Ignite keynote on 17 November. “That’s 180x the current throughput,” he said.

By our arithmetic, 180x implies a starting point of about 40 writes a second. GitHub’s own engineers wrote on 5 November that its busiest repositories today receive “at most dozens of writes in a single second.”

GitHub is owned by Microsoft. Our Microsoft stock page tracks the company’s price and key dates.

Business Pill · The Model and the Harness

An evergreen Business Engineer explainer. It teaches the general idea only and does not discuss this report.

The key insight: Nadella said on 17 November that GitHub’s rebuilt Git storage sustained 7,200 writes a second from 5,000 agents on one repository in testing, ‘180x the current throughput’; GitHub says busiest repositories now see at most dozens of writes a second.

What Nadella Said

“GitHub is seeing exponential increases, the pushes, the commits, the actions, the PRs, all being driven by agents,” Nadella said, according to Microsoft’s transcript. “And we’re rebuilding; that means all of GitHub’s core, starting with the Git architecture.”

He described the test this way: “In testing we pointed, I think, 5,000 agents to one repo, and we’re able to sustain 7,200 writes per second. That’s 180x the current throughput. That’s the improvement and scale we need for an agentic workload.”

Nadella said GitHub sustained 7,200 writes a second with 5,000 agents on one repository in testing, '180x the
Nadella said GitHub sustained 7,200 writes a second with 5,000 agents on one repository in testing, ‘180x the current throughput’; that implies about 40 a second today, by our arithmetic.

What GitHub Rebuilt

A repository is the shared folder where a software project’s code and its history live. Every time a person or an agent saves work to it, that is a write, or a push.

GitHub detailed the new design, called Rubicon, on its engineering blog on 5 November. Code data moves to Azure Blob Storage, while the pointers to each branch live in a separate service GitHub calls Ref-CAS. The aim is that a push needs as little coordination as possible.

GitHub says it is available in a limited private preview, with a broader rollout early next year.

Why Agents Broke the Old Design

“An agent in a tight loop commits or checkpoints after nearly every action,” GitHub wrote on 6 October, so delays “a human would never notice” become the limit. Thousands of agents working in one repository create a steady stream of writes, it said.

GitHub said pushes grew 4.9x in a year, from 0.69 billion to 3.35 billion a month. Commits reached 7.38 billion in September, more than five times as many as a year earlier, and total Git activity more than doubled, it said.

Bar chart: GitHub commits up more than 5x, pushes up 4.9x and Git events up 2.2x year on year
GitHub says pushes grew 4.9x in a year and September commits were more than five times a year earlier.

Two Numbers, Two Tests

GitHub’s 6 October post put the gain at “up to 35 times higher write throughput” in internal benchmarks. Nadella’s 180x figure came six weeks later and describes a different test, 5,000 agents on one repository.

The Structural Read

As we read it, the constraint on coding agents is moving from the model to the plumbing that stores their work.

As we read it, the 180x and GitHub’s earlier 35x describe different tests, so the production rollout is the number that will settle it.

Satya Nadella, Microsoft Ignite keynote, 17 Nov 2026

“That’s 180x the current throughput. That’s the improvement and scale we need for an agentic workload.”

Three Implications

FOR ENGINEERING TEAMS GitHub says Rubicon is in limited private preview, with a broader rollout early next year.

FOR AGENT BUILDERS GitHub says its busiest repositories see at most dozens of writes a second today.

FOR MICROSOFT Nadella said what worked for GitHub now needs to be pervasive across Azure.

What Is Not Established

Neither GitHub nor Microsoft has published the full test set-up, so the two multiples should not be compared directly. Both describe internal tests, not production traffic, and Nadella hedged the agent count with “I think”.

The Wider Pattern

It fits other signs that agents are now a large share of software traffic. We reported that agents now start over 40% of GitHub pull requests, that agent pull requests merge at about 90%, and that Stripe saw AI agents’ spending rise nearly 80x since September.

This Week In The AI Supercycle

A World Without an Order, and an AI Central Bank Without a Successor

This week’s issue: the week AI memory stayed largely sold out while the people holding the shares paid: why the AI Supercycle is not a dot-com replay, read through geopolitics, macro and technology, with the gap between building and financing the widest we’ve recorded.

Read the full issue →

Business Engineer Framework

The AI Supercycle

The Business Engineer map of where AI capability turns into business leverage.

Read the Map of AI →

The Bottom Line

Nadella says GitHub’s rebuilt Git storage sustained 7,200 writes a second from 5,000 agents in testing, which he called 180x the current throughput. What to watch is the broader rollout GitHub promises for early next year, and whether production numbers match the tests. This is not investment advice.

97,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.

A note on sourcing. Nadella’s words are from Microsoft’s official transcript of his 17 November Ignite keynote, which The Transcript recirculated on X on 22 November. GitHub’s design details and growth figures are from its engineering blog posts of 6 October and 5 November. The roughly 40-writes-a-second baseline and the 2.2x activity multiple are our arithmetic. Both throughput figures describe internal tests. This is not investment advice.

Sources: Microsoft, Satya Nadella, Microsoft Ignite 2026 opening keynote transcript, 17 Nov 2026 · GitHub Blog, ‘Inside GitHub’s new Git storage architecture’, 5 Nov 2026 · GitHub Blog, ‘Building Git infrastructure for agent-scale development’, 6 Oct 2026 · The Transcript on X, quoting the keynote, 22 Nov 2026 · FourWeekMBA on agents starting over 40% of GitHub pull requests, 8 Nov 2026 · FourWeekMBA on agent pull requests merging at 90%, 29 Oct 2026 · FourWeekMBA on Stripe’s agent spending data, 21 Nov 2026

Scroll to Top
FourWeekMBA