The whiteboard had been in Justin's office since the factory was founded. It had gone through phases. In July, it was covered in diagrams—boxes and arrows, the visual grammar of systems thinking. In August, the diagrams gave way to equations. In September, the equations simplified. In October, they simplified further. By November, the whiteboard held a single line of notation that neither Jay nor Navan could fully parse.
In December, Justin erased everything and started over. He stood at the whiteboard for three hours on a Saturday, alone in the office, and when Jay arrived on Monday there were three terms connected by two operators, and the whiteboard was clean except for that one expression.
By January 2026, it had stabilized. The equation hadn't changed in six weeks. Three terms. Two operators. No extraneous notation. Justin had circled it once, in blue marker, and written "stable" underneath.
Jay had been avoiding it. Not consciously—he walked past it every day—but he hadn't asked about it since November, when Justin's explanation had used words like "basin" and "manifold" and "invariant set" and Jay had nodded along without genuine comprehension. But in January, something shifted. The factory had been running long enough that Jay's intuition about its behavior had deepened. He understood, not intellectually but experientially, how satisfaction metrics behaved, how agent convergence worked, how scenarios and specifications interacted to produce reliable software.
He looked at the equation on a Tuesday morning. And he understood it.
The first term described the relationship between specification clarity and agent convergence rate. Higher clarity, faster convergence, but with diminishing returns past a threshold. The second term described the role of scenario diversity as a forcing function—more diverse scenarios expanded the attractor basin, increasing the probability that the system would converge to the correct solution rather than a local minimum. The third term was the cost function: tokens, time, compute, normalized against satisfaction improvement.
Three terms. The whole factory, expressed as a mathematical relationship between clarity, diversity, and cost.
"I understand the whiteboard," Jay said to Navan during lunch.
Navan looked at him. "So do I. Since last week."
"You didn't say anything."
"I was hoping I was wrong."
They looked at each other. Understanding the equation meant understanding the implications. It meant understanding that the factory's behavior was not just empirically observable but mathematically predictable. It meant that the system they had built was not a collection of clever hacks but a genuine dynamical system with describable properties. It meant that what Justin had seen from the beginning—the deep structure beneath the surface complexity—was real.
"He's been right the whole time," Jay said.
"About the math, yes," Navan said. "About the implications..." He trailed off. He opened his notebook. He wrote something and closed it. Jay did not ask what he had written.
Justin found them staring at the whiteboard after lunch. He looked at their faces and understood immediately. "You see it now," he said. It was not a question.
"We wish we didn't," Jay said.
Justin smiled. It was the smile of someone who had been carrying a weight alone and had just felt others take hold of it. "I know," he said. "I know."
"We wish we didn't" is doing so much work in this story. Understanding the math means understanding that this isn't going to stop. This is what software becomes.