What You’ll Learn

The title is a claim and James defends it directly: the agent that changes your business isn’t a better specialist, it’s the one that runs the others.

This session covers why orchestration beats adding agents, how to stop overpaying by routing work to the right model, and the argument that your accumulated context is a business asset rather than a byproduct.

The Core Claim

“The goal is not to build agents. The goal is to build an operating system with an orchestration officer.”

The practical difference shows in how you talk to it. James doesn’t ask how to do anything — he asks for outcomes. “I want fifteen leads.” “I want daily shorts.” Routing to the right department and the right employee is not his job.

💡 In Plain English: One more specialist agent gives you one more thing to remember. An orchestrator gives you one thing to talk to.

Model Routing: Stop Paying Premium for Routine Work

A genuinely practical piece. Not every step needs the most capable model — and running everything on the top tier burns through a plan fast.

The routing policy sends bulky, mechanical steps to a lighter model while judgment, voice and anything customer-facing stay on the strong one. James also runs a local model on a Mac Mini for tasks that don’t need to leave the house.

✓ Check Your Work: List your recurring AI tasks and mark each mechanical or judgment. The mechanical column is what you’re currently overpaying for.

⚠️ Keep the pins deliberate: anything carrying your voice, anything that publishes, anything touching money, and anything needing your judgment should stay on the strong model regardless of cost.

Context as a Sellable Asset

The idea James is most animated about. Everything you’ve published, taught, answered and curated — transcripts, tutorials, posts, courses — is a body of context nobody else has.

Put it in a vector database and your agents can draw on it. That’s the difference between output that could have come from anyone and output that could only have come from you.

It also reframes community work. The archive of what you’ve said and how you’ve said it isn’t exhaust. It’s the training material for your own staff, and it compounds.

The Platform Argument

His view on what a durable education business needs: a platform you own, a content library, and live events. Not social media alone, because you don’t own the audience or the archive there.

Building Employees From Repetition

A neat mechanism worth copying even without this system. The orchestrator watches what you keep asking for.

“Hey, we’ve been doing this three times or four times over the last little while. Maybe we should build an employee for that.”

Repetition is the signal. Rather than deciding up front what agents you need, you let the actual work tell you — then approve the hire.

The Bottleneck

“Remove yourself from the bottleneck. You don’t want to be the bottleneck other than the ideas that you have.”

The Michael Gerber framing sits underneath this: most people running a small business are technicians doing the work, not entrepreneurs directing it. AI makes the technician faster, which feels like progress and isn’t.

Where He Thinks It Goes

His stated prediction: 2026 is agents and avatars, 2027 is conversations — talking to AI employees to get real work done rather than composing prompts. Treat that as a forecast he’s putting his name to, not a fact.

Key Takeaways

• Ask an orchestrator for outcomes rather than adding another specialist agent you have to remember and manage.
• Route mechanical work to cheaper or local models and pin voice, publishing, money and judgment to the strong one.
• Your published archive is context nobody else has — put it somewhere your agents can query it.
• A durable education business needs an owned platform, a content library and live events, not social reach alone.
• Let repetition tell you which AI employee to build next instead of guessing the roster up front.

Your Next Step

Count how many times you’ve asked for the same kind of output in the last month. Three or more and that’s your next hire.

The Campus AI OS is free: trainingsites.io/os