How to run a company in 12 repositories

​Rob Prickett asked me why I split company work across repositories instead of putting everything in one place.

​The answer has surprisingly little to do with code. Product, relationships, market intelligence, investor work, company operations, and agent tooling each need different context, instructions, automation, history, access, and review.

​So we are turning the conversation into a live office hour.

​I will open the real operating system behind the company and trace one piece of work across its seams. Rob will ask the questions he has been asking in private:

​How does an agent work across repositories? Where does shared context live? What stays in a local runbook, and what becomes a reusable skill? How do permissions change the architecture? What has broken, gone stale, or cost more to maintain than expected?

​This is for people already using Codex, Claude Code, Cursor, or an internal agent harness in real work. It is not an AI introduction, a generic tool tour, or a product demo.

​Bring one question about context, repository seams, permissions, or reusable workflows. We will leave time for the room. You will also get the seam-contract worksheet used in the teardown so you can map one workflow in your own system.