An interviewer once asked Justin how many engineers worked on the factory. Justin said three. The interviewer assumed he meant three hundred, or three teams of engineers, or three principal engineers leading larger groups. Justin let the silence stretch until the interviewer understood.
Three. Three people.
Seven months. Four open-source projects. Six digital twins, soon to be seven. Thousands of scenarios. Over 50,000 agent-authored commits across four repositories. A Rust server with 16,000 lines. A Go gateway with 9,500 lines. A TypeScript frontend with 6,700 lines. A pipeline runner, a convergence engine, a container security framework, and an identity service. A satisfaction metric approaching 0.99.
Three people.
The math didn't add up. That was what everybody said, the first time they heard it. The math doesn't add up. Three people can't produce that volume of code. Three people can't maintain four repositories in three programming languages. Three people can't run thousands of scenarios per hour across six digital twins.
And they were right. Three people couldn't do those things. Three people didn't do those things. Three people wrote scenarios, defined goals, set policies, and described the world they wanted to exist. The agents did everything else.
The math didn't add up because the math was wrong. The traditional equation—output equals engineers times productivity—assumed that output was bounded by the number of humans in the room. The factory's equation was different. Output was bounded by the quality and quantity of scenarios. Humans wrote scenarios. Agents wrote everything else. The humans were the bottleneck in one dimension—intention, direction, judgment—and the agents were unbounded in the other.
Navan tried to explain this to his parents during a weekend call. "I don't write code," he said. "I describe what the code should do." His mother said that sounded like management. His father said it sounded like architecture. Navan said it was neither and both and something new that didn't have a name yet.
Jay tried to explain it to a friend from a previous job. The friend asked how many pull requests Jay reviewed per week. Jay said zero. The friend asked how many he wrote. Jay said zero. The friend asked what he did all day. Jay said he wrote scenarios and watched dashboards and thought about edge cases. The friend changed the subject.
Justin didn't explain it to anyone. He'd published the paper. The paper said everything he wanted to say. If people didn't believe three people could do this, the code was open-source. They could look.
Three people. A room. Laptops. Coffee. A dashboard on the wall. And a factory that ran day and night, writing code that no human wrote, reviewed, or read, producing software that worked better than anything a hundred-person team could build.
The math didn't add up.
That was the point.
Navan's parents calling it management and architecture. Jay's friend changing the subject. These are the real moments—the ones where the people doing the unprecedented thing try to explain it to people living in the precedented world.