The org chart is a codebase.
How one human plus an AI team actually runs the work: roles with charters, processes written as SOPs, shared context — and the places where it still breaks.
Two layers, one company.
The decision layer is one person. They choose what is worth building, write the specs agents execute against, verify outcomes, and own every result — including the ones agents got wrong.
The execution layer is the AI team. Agents organized by function — engineering, data, commerce, research, growth — each with a written charter, its own tools and an SOP. They ship code, sync catalogs, fulfill orders, run experiments and draft what you are reading now.
Nothing about this lives in someone's head. Roles, decisions and postmortems are versioned artifacts the next agent executes against. That is what makes the model transferable — a playbook in files can be handed to the next person; a playbook in a head cannot.
Judgment doesn't delegate
- Agents produce volume, not conviction. Every wrong spec gets executed faithfully, at scale.
- The bottleneck moved from "can we build it" to "did we describe the right thing".
Verification is the new labor
- Output that looks right is cheap and abundant. Output that is right still needs a human or a test.
- Most of the human's time goes to specs and acceptance, not creation.
Context is the real infrastructure
- An agent without the company's context is just an API call.
- Shared memory — one product catalog, decisions in writing, versioned SOPs — is what makes the team coherent.
Today — generalist agents execute
One human directs; AI agents already ship code, run operations and keep the storefront alive around the clock.
Next — role-specialized agents
Agents organized by function — product, engineering, growth, operations — each with its own charter, tools and SOP.
Horizon — self-running functions
Whole departments operating autonomously under human review, with escalation paths instead of micromanagement.
Constant — the human decides
What changes is how much the AI team carries. What doesn't: a person chooses what is worth doing, and answers for it.