What You’ll Learn

A long working session where James shows a morning’s completed work, then explains the architecture behind it. The most portable idea in it changes how you choose software: pick tools for their connectors, not their features.

Connectors Beat Features

This is the practical takeaway even if you never install anything he sells.

A tool’s feature list tells you what you can do inside it. Its connector tells you what your agents can do with it. Once most of your work is delegated, the second matters far more than the first.

💡 In Plain English: A connector is the doorway between your AI and your software. A brilliant tool with no doorway is a tool you still have to operate by hand.

✓ Check Your Work: List the five tools you use most. For each, can your AI read from and write to it directly? The ones that can’t are where your manual work will pile up.

Outcomes, Not Instructions

“Not what tools are you using, it’s who, what staff do you have?”

The work he shows was produced from natural-language requests for outcomes rather than step lists — a reminder email, sales pages updated, a quick-start guide, an announcement, slides. What makes that possible isn’t a better prompt; it’s departments that already know the business.

Playbooks vs Workflows, Again

Worth repeating because it’s the distinction most people collapse.

A workflow is a procedure inside one area — the steps of writing a blog post.

A playbook is a chain of outputs across departments: video → transcript → content brief → article → community post → social → email. Playbooks can be written by hand or created automatically once the same sequence has run a few times.

Closed-Loop Tracking

Every job carries an identifier that connects who started it, which employees worked it, what each produced, and what happened at the end. Without that you can’t answer “did that work?”, which means you can’t improve anything deliberately — you’re just running things and hoping.

Model-Agnostic by Design

The system now runs across multiple assistants and local models rather than being tied to one vendor. Given how fast the frontier moves, being locked to a single provider is a risk rather than a simplification.

⚠️ James makes several forecasts in this session — on release cadence, on voice arriving, on the skills layer losing value. They’re his reads, not established facts, and worth weighing as such.

Talk to It Like a Person

A small note he returns to at the end and it’s the one most people get wrong. He talks to Dean rather than issuing commands, and brings a council in to pressure-test an idea before work starts.

“I call him Dean because I want one person that I can talk to in my business.”

The strategy conversation happens first. The production happens after — and only once there’s an agreed angle.

Key Takeaways

• Choose software by whether your AI can connect to it, not by its feature list — the connector decides how much can be delegated.
• Ask for outcomes; the department structure and shared memory are what let a plain request turn into finished work.
• A playbook chains outputs across departments; a workflow is the procedure inside one of them.
• Give every job a tracking identifier, or you can’t tell what worked and can’t improve deliberately.
• Stay model-agnostic — locking to one provider is a risk while the frontier moves this fast.

Your Next Step

Audit your stack for connectors before you buy anything else. The tool you’re about to switch to may be worse at the job and far better at being delegated — and that’s now the more valuable property.

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