Can you spin up a whole AI company for one person?

Hypothesis

If an autonomous agent organisation works for us, it should be packageable: a self-contained "pod" — chief of staff, specialist employees, its own messaging front door, optionally its own computer — provisioned for one client, sealed from every other tenant. One person gets a whole staffed company.

What we built

The pod pattern, now running for two early cases: a sealed container per client holding a full agent org (personas defined in plain-text soul files, work dispatched through a kanban board), a dedicated Telegram bot as the client's front door, and access to a shared virtual-desktop driver for screen work. Internal pods may connect to our OS; client pods are sealed with egress allow-lists. The same substrate serves both — one flag flips a pod from internal to client-grade.

Learnings

• Persona files are the product. A pod's usefulness tracks the quality of its written identities and their delegation contracts far more than the underlying model.
• Multi-tenancy is a trust boundary, not a feature: backups, approval gates, and injection-resistance drills had to be backported to every pod the week we hardened the mothership.
• Every client teaches the framework — pod setup is a documented, repeatable recipe now, with a fleet-learnings log appended every time an org change proves out.

Log

• 2026-07-11 — Both pods hardened: approval gates, verified backups, untrusted-input guards across all personas.
• 2026-07-09 — Pod pattern defined; first two pods provisioned on a dedicated server.