The argument started over lunch and lasted through dinner.
The Digital Twin Universe had six twins: Okta, Jira, Slack, Google Docs, Google Drive, Google Sheets. Six services, six behavioral clones. The question on the table—literally on the table, next to the takeout containers—was whether to add a seventh.
"GitHub," Navan said immediately. "It's not even a question. Half our agent workflows involve Git operations. The agents push code, open pull requests, run CI pipelines. We're testing all of that against real GitHub right now, which means rate limits, API costs, and leftover test repositories cluttering our organization."
"Linear," Jay countered. "Jira is the issue tracker we have a twin for, but a lot of teams are moving to Linear. If we're building the factory as a product, not just an internal tool, we need twins for the services customers actually use. And Linear's API is cleaner than Jira's. The twin would be easier to build."
"Easier to build isn't a good enough reason to build it," Justin said. He was eating a salad and had been listening more than talking, which was usually a sign that he had already formed an opinion and was waiting for the right moment to share it.
"Notion," Navan offered as a second candidate. "It's becoming the default documentation layer for a lot of companies. Pages, databases, blocks, permissions. It's a sprawling API but a useful one."
"The question isn't which service," Justin said, setting down his fork. "The question is what scenarios are blocked by the absence of a seventh twin. Which scenarios can't run today because we don't have a twin for something?"
Silence. Jay pulled up the scenario catalog on his laptop. He filtered for scenarios that interacted with services outside the DTU. There were fourteen. Nine of them touched GitHub. Three touched AWS. Two touched Datadog.
"Nine scenarios blocked on GitHub," Jay reported.
"And how many of those nine are in the critical path for satisfaction scoring?"
Jay checked. "Four."
"Okay. So GitHub has a concrete impact on our ability to measure satisfaction across four critical scenarios." Justin picked up his fork again. "Linear and Notion have zero scenarios waiting on them. They're solutions looking for problems."
Navan conceded the point but not the argument. "Linear and Notion are forward-looking. We'll need them eventually. Why not start now while we have bandwidth?"
"Because twin maintenance is not free," Justin said. "Every twin we build is a twin we have to keep current. API changes, behavioral drift, SDK compatibility—it's ongoing work. Six twins is already a significant maintenance burden. Seven is doable. Eight starts to hurt. We add twins when scenarios demand them, not when we have spare cycles."
"So GitHub," Jay said.
"Not yet. I want you both to think about it for a week. Write up the scope. How many API endpoints. How complex the behavioral model. What the fidelity targets would be. What SDK compatibility looks like. Then we'll decide."
Navan closed his laptop. Jay closed his. Justin finished his salad.
They tabled it for a week. The week would turn into two, as these things do. But the question had been asked, and questions in the factory had a way of getting answered, sooner or later, by the convergent pressure of scenarios that needed to run.
Justin asking "what scenarios are blocked" instead of "what would be cool" is such a grounding question. Deciding by demand rather than desire. Good leadership moment.