What You’ll Learn

James opens this session with a line worth repeating: he has an agent that asked to hire another employee. That only becomes possible once your agents stop being a pile of tools and start being an organisation.

This walks through what an agent org chart is, why a small business needs one, and the six capabilities that only exist once you have one.

Roles, Not Tools

The core reframe is simple and it changes everything downstream.

“You’re hiring roles, you’re not hiring tools or features. You’re not hiring skills, you’re actually hiring business roles.”

James’s own chart has a chief orchestration officer — Dean — sitting above six departments, with a strategy and research council alongside. But the shape is not the point. An education business, a consultancy and a home services business need different departments, and the chart should match the business you actually run.

💡 In Plain English: An org chart is just a written answer to “who does what, and who hands off to whom.” Your agents need that answer as much as human staff do.

Why an Orchestrator Sits on Top

Without someone conducting, you are the one delegating every step of every multi-step job. James is explicit that he does not want to know his employees by name or track what each one does.

What he wants from an orchestrator is narrower and more useful: decide whether a job is worth doing, route it to the right department, confirm the work was done properly, and hold the handoff accountable.

The Six Capabilities

1. A team directory

The orchestrator knows how many AI employees exist, what department each sits in, and what each can do. You stop maintaining that list in your head. If you have fifteen skills and can’t remember which does what, this is the fix.

2. Playbooks

Playbooks run across multiple departments and multiple agents. Some ship with the system, some you write, and some get created automatically as you work.

3. It hires its own staff

When the orchestrator hits a job with no employee to hand it to, it notices the gap and proposes a hire. Crucially it does not create one on its own — it drafts the candidate and brings it to you.

4. Department managers

A manager holds the memory, goals and tasks for one department. Standing instructions like “this is how we do social” or “this is how we write blog posts” live with the department, not smeared across every employee.

5. End-to-end routing

This is the one James says nobody is talking about. Every job carries a run ID that connects the agent, the output and the outcome. You can ask what happened, who did it, and what came out.

“A staff you can’t audit isn’t a staff.”

6. Playbook routing for compound jobs

“Launch this product.” “Run this cohort.” These cross marketing, sales and delivery at once. With a playbook the orchestrator already knows the sequence and just runs it.

Playbook vs Workflow — Not the Same Thing

James draws a distinction worth holding onto.

A workflow is a standard operating procedure — open this app, choose this menu, write the copy, check the reading level. It’s the shape of a task.

A playbook is written so a system can execute it. It has named triggers, ordered stages, and one owner per stage. It adds quality checks between stages, human stop points, and limits on how far it runs unsupervised.

✓ Check Your Work: Take a job you repeat monthly. Can you name its trigger, list its stages in order, and name one owner per stage? If yes, that’s a playbook. If it’s a list of clicks, it’s a workflow.

What Happens Without One

The failure mode is improvisation. Ask for a compound outcome with no playbook and the model fills the gaps itself, which means you get a different result each time and no way to tell why.

⚠️ Worth knowing: Repeatability is the payoff here, not speed. Once you have watched a playbook run a couple of times and you’re happy with the output, you stop needing to be there for it.

Key Takeaways

• You are hiring business roles, not tools, features or skills — the org chart is the hiring plan.
• An orchestrator’s job is deciding what’s worth doing and holding the handoff accountable, not doing more work.
• A run ID on every job is what makes agents auditable; without it you cannot tell what produced a result.
• Keep department memory with the department, so standing rules don’t have to live in every employee.
• Write playbooks with named triggers, ordered stages and one owner per stage, or the model improvises and results stop being repeatable.

Your Next Step

Sketch your own chart before you install anything. What are the departments in your business, and which one is completely unstaffed right now? That gap is where your first AI hire goes.

The Campus AI OS is free and ships with an orchestrator and a starting roster: trainingsites.io/os