Git Was Built for Humans. Agents Are Rewriting It.

GitHub just published how it is rebuilding Git infrastructure for agent-scale development. Read it next to Entire, the new company from former GitHub CEO Thomas Dohmke, and Cursor Origin, and one thing is clear: Git was built for humans, and agents are now rewriting it.

I have spent nine years building analytics on top of Git. Waydev even holds a US patent on Git analytics. So I want to share what I think these three moves mean, and what is still missing.

The numbers behind the rebuild

GitHub’s post is an infrastructure deep dive, but the data is the story. Developers and agents made 7.38 billion commits on GitHub in September alone. Total Git activity more than doubled in a year. The busiest single repository handled roughly a billion requests in August.

The detail that stayed with me is about rhythm. An agent in a tight loop commits or checkpoints after nearly every action, so push latency a human would never notice becomes the agent’s speed limit. Thousands of agents on their own branches in one repository create a write load the old architecture was never designed for. As GitHub puts it: “Writes are way harder.”

Their answer is to decouple storage from compute (moving beyond Spokes, the system that stores every repository today), coordinate only the reference update that truly needs agreement, and move maintenance off the serving path. In internal benchmarks the new architecture has delivered up to 35 times higher write throughput. And one of their stated principles is to keep people in control of their code, so the people who own it can still review, understand and approve what agents produce. That is the same separation of intelligence and authority I wrote about in Models as Insider Risks in the Super Intelligence Era.

Three bets on Git for agents

All three are right about something. And all three describe the same shift: the commit is no longer a unit of human intent. It is a checkpoint in a machine loop.

Why this breaks how we measure engineering

For years, a lot of engineering analytics, ours included, started with Git activity: commits, pull requests, lines changed. Those signals made sense when every commit came from a person with a plan. When an agent commits after every action, activity tells you almost nothing. A 5x jump in commits is not a 5x jump in value. It might even be a 5x jump in rework.

The questions leaders need answered move up a level. Not how many commits, but which agent produced the change, in which tool, at what cost, who approved it, and whether it held up in production. This is the shift I described in The SI SDLC, and why your ability to trust agent output has a limit. Entire answers part of the first question. GitHub and Origin will host the work. Nobody in that picture is built to answer the last one across all of them.

My view: enterprises will be multi-agent and multi-forge

Large enterprises will not pick one agent or one forge. They will run several coding agents side by side, keep code across more than one platform, and add tools like Origin or Entire next to GitHub, not instead of it. That makes a neutral outcome layer more important, not less. It has to sit outside the agents and outside the forges, exactly as Satya argued the controls must sit outside the model. Many of the engineering leaders building the agent stack are wrestling with exactly this.

That is what we are building at Waydev. We measure AI adoption, impact and ROI across humans and agents, whatever tool wrote the code and wherever it lives. And with Waydev 15B, our own SDLC model, it runs tailored to each customer, private and air-gapped inside their environment. When your Git history becomes a machine log, the analysis of it should stay inside your walls too.

Here is how that maps to the outcome layer in practice:

Agents made commits cheap. The question that matters now is which ones were worth making.

Alex Circei, CEO and Co-Founder of Waydev

If you are rethinking how you measure engineering for the agent era, our guide to measuring AI adoption, impact and ROI and the Impact That Compounds playbook lay out the metrics that still mean something. For agents specifically, start with the How to Measure AI Agents Playbook or run your numbers through the AI ROI Calculator.