The question came from a board member. It was the kind of question board members ask because they are constitutionally obligated to think about continuity. "What's the succession plan for the factory team?"
Justin had been expecting the question. He had been expecting it because it was the wrong question, and wrong questions arrived on predictable schedules.
"There isn't one," he said.
The board member—a reasonable person, experienced in technology companies, accustomed to hearing things he didn't like—blinked. "There has to be a succession plan. What if one of you leaves? What if all of you leave?"
"Then the factory keeps running."
"Who runs it?"
"Nobody. It runs itself."
The silence that followed was the specific silence of someone reassessing their understanding of a system they thought they understood. Justin let it breathe. He had learned, over years of explaining technical concepts to non-technical decision-makers, that the gap between what was said and what was heard was often wider than the gap between what existed and what was believed possible.
"The factory is a convergent system," Justin continued. "The specifications describe the desired behavior. The scenarios validate that behavior. The agents produce and iterate on code until the satisfaction metric meets the threshold. The pipeline runs continuously. None of those components require human intervention during operation. Our role is to improve the specifications, expand the scenarios, and refine the twins. If we stopped doing that, the factory would continue to produce software at its current level of quality. It just wouldn't improve."
"So you're maintenance," the board member said.
"We're gardeners," Justin said. "We tend the system. We don't run it. If we left, the garden would still grow. It would just grow without direction."
Jay, who was present for this meeting in a support capacity, resisted the urge to smile. Justin had used a gardening metaphor. The man literally gardened in his free time. The metaphor was not accidental.
"But eventually someone has to decide what to build next," the board member pressed. "Someone has to write new specifications. Someone has to choose which problems to solve."
"Yes," Justin said. "And anyone can do that. The specifications are in natural language. The scenarios are in natural language. The methodology is open source and documented. A new person could learn to operate the factory in weeks. A new person already has—that was the Berlin team's experience. Two weeks from zero to functional."
The board member wrote something in his notebook. Jay couldn't see what it was, but the handwriting was fast and small, the handwriting of someone updating a mental model.
"So the succession plan," the board member said slowly, "is that the factory doesn't need successors."
"The succession plan is that the factory was designed from the beginning to outlive its creators. Not as a goal. As a property. A system that depends on specific people to function is a system with a single point of failure. We don't build single points of failure."
The board member closed his notebook. "That's either the best succession plan I've ever heard or the most terrifying."
"Those aren't mutually exclusive," Justin said.
The bus factor of the factory is zero. Not because the team is irreplaceable, but because the team is unnecessary for continued operation. That's a sentence I never thought I'd type about software development.