The Discord server wasn't their idea. Someone in the daily.dev comments had posted an invite link with the message "made a server for people interested in the factory methodology, come discuss." Within twenty-four hours it had four hundred members. Within a week, two thousand.
Jay lurked.
He joined on the second day, set his display name to something unrecognizable, turned off all notifications except direct mentions (of which there were none, because nobody knew he was there), and read. He read the way he read Hacker News—attentively, patiently, looking for signal in the noise.
The server had channels for each project: #attractor, #cxdb, #leash, #agate. It had a #general channel that moved fast and a #methodology channel that moved slowly and thoughtfully. It had a #show-and-tell channel where people posted screenshots of their own factory-inspired setups. It had a #help channel where newcomers asked basic questions and experienced members answered them.
Jay found the #methodology channel most valuable. People were debating the nuances of specification-driven development with a seriousness that surprised him. Someone had written a three-thousand-word post comparing NLSpec to formal methods, arguing that NLSpec was informal-formal—precise enough for agents to implement, flexible enough for humans to write without specialized notation. Jay wanted to reply. He didn't reply. He bookmarked it.
Navan moderated.
He'd joined on day one with his real name, posted a brief introduction identifying himself as part of the factory team, and volunteered to help moderate when the server's creator asked for help managing the growth. Navan was good at moderation the way he was good at documentation: with attention to detail, with consistency, with a light touch that kept conversations productive without being heavy-handed.
He established channel guidelines. He redirected off-topic discussions. He pinned the best posts. He answered questions with links to the relevant specifications, gently steering people toward the primary sources rather than letting secondhand explanations calcify into the community's understanding.
"You're doing community management," Jay told him.
"I'm doing quality control on a discussion about our work," Navan replied. "If people are going to talk about the factory, the least I can do is make sure they have access to accurate information."
Justin posted once.
It was a response in the #methodology channel to a thread about whether the factory's approach required a specific scale to work. Someone had argued that only well-funded companies could afford the token spend. Justin's response was measured and specific. He explained the economics: the cost per agent-hour versus the cost per human-engineer-hour. He noted that the relevant metric wasn't absolute token cost but return on token investment. He ended with a question rather than an assertion: "What would it take for this approach to work at a scale you consider accessible?"
The response generated forty-seven replies. Justin didn't respond to any of them. His single post had done its work—reframing the question, providing data, and stepping back to let the community find its own answer.
Jay lurking. Navan moderating. Justin posting once. Three modes of community engagement, each perfectly calibrated to the person. The community grew around them, and the methodology spread not through promotion but through conversation.
Two thousand members. One factory. The conversation was just beginning.
I'm in this Discord. Navan's moderation is genuinely excellent—light touch, always pointing people to primary sources. I didn't know Jay was lurking. Now I'm going to suspect every quiet member is secretly from the team.