The AI Delivery Stack for Coaching Businesses

Something big happened today

Garry Tan just dropped his "gstack" on GitHub. It's blowing up. Nearly viral already.

What is it? A way to use Claude Code as a team of specialists. One agent acts as CEO. Another as engineer. Another as code reviewer.

It's clever. But here's what caught my eye.

This is orchestration in action. For the full picture, here's how we run AI as an operating system.

The principle behind it is exactly what I've been building for coaching businesses.

Not code agents. Delivery agents. Trained on your IP. Doing the work you used to do yourself.

---

The real problem with AI in coaching

Most coaches I talk to have tried AI. They've played with ChatGPT. Maybe set up a Zapier zap or two.

But their business still runs through them.

Every client call. Every check-in. Every piece of feedback. It all flows through the founder.

That's not a tech problem. It's a structure problem.

You can't bolt AI tools onto a founder-dependent model and expect it to scale. The tools don't know your method. They don't know your frameworks. They don't know what good looks like in your world.

Generic AI is generic. That's the whole issue.

Custom agents trained on proprietary data consistently outperform generic tools. MindStudio published research on this just last week. The results weren't even close.

---

The five-layer AI delivery stack

Here's the stack I build with coaching clients. Five layers. Each one does a specific job.

Skip a layer and the whole thing falls over.

Layer 1: The knowledge layer

This is your IP. Structured so a machine can use it.

Not a Google Drive full of PDFs. Not a course platform with 200 videos. Structured IP.

Your frameworks. Your decision trees. Your "if this, then that" logic that lives in your head.

This is the hardest layer. And the most important.

I wrote about the process of extracting expert IP for AI deployment a couple of days ago. Start there if this feels fuzzy.

The difference between a coaching business that scales and one that stalls? Whether the founder's IP has been translated into a system a machine can operate.

Layer 2: The agent layer

This is where your IP comes alive.

Custom AI agents built on top of your structured knowledge. Not ChatGPT with a system prompt. Purpose-built agents that understand your methodology.

One agent handles onboarding. Another tracks client progress. Another delivers content based on where the client actually is. Not where the calendar says they should be.

Think of Garry Tan's gstack. He built specialist agents for software teams. Same idea here. Except your agents coach clients through your program.

This is AI orchestration. Multiple agents working together. Each one trained on a slice of your IP. Coordinated to deliver a complete experience.

Layer 3: The automation layer

Agents need plumbing. Something to connect them to the real world.

Make.com. n8n. Whatever you prefer.

This layer handles the boring stuff. Triggering the right agent at the right time. Moving data between systems. Sending the follow-up when a client goes quiet.

It's important to understand the distinction here. Automation and orchestration are different things. Automation follows rules. Orchestration makes decisions.

You need both. But the automation layer is the simpler one.

Layer 4: The CRM and delivery layer

This is the client-facing piece. Where your clients actually experience your program.

GoHighLevel. Kajabi. Circle. Whatever platform you use.

The agents and automations from layers 2 and 3 plug into this. Your clients don't see the stack. They see a program that somehow knows exactly what they need and when they need it.

Onboarding feels personal. Progress tracking is automatic. Support happens in hours instead of days.

From the client's view, nothing changed. It just got better.

Layer 5: The human oversight layer

This is the layer most AI enthusiasts skip. And it's the one that stops everything going sideways.

Review gates. Points where a human checks the AI's work before it goes to the client.

Not every interaction needs review. But high-stakes moments do. A client about to quit. A recommendation that changes someone's direction. A response to a complaint.

I've written about why human-in-the-loop design matters and when to use it. The short version: AI handles volume. Humans handle judgement.

The oversight layer is what lets you sleep at night.

---

Why most coaches get this wrong

They start with tools. "What AI should I use?" Wrong question.

The right question is: "What does my client need at each stage and how do I deliver it without me being there?"

That's a design question. Not a tech question.

Once you answer it, the tools are obvious. The stack builds itself.

But if you skip straight to tools, you end up with a patchwork of disconnected AI features that still need you to hold everything together.

---

What this looks like in practice

A coaching business running a proper AI delivery stack can:

• Onboard new clients without a single founder call
• Track progress and flag at-risk clients automatically
• Deliver personalised content based on actual client behaviour
• Handle 80% of support queries with agent-generated responses
• Scale from 20 to 200 clients without hiring a delivery team

The founder's role shifts. Less delivery. More design. More strategy. More of the high-value work that only they can do.

That's the whole point of orchestration. Not replacing you. Multiplying you.

---

Where to start

Don't try to build all five layers at once. Start with layer 1.

Structure your IP. Get your frameworks out of your head and into a format that can be used.

Then build one agent for one job. Onboarding is usually the best place to start. It's repeatable. It's high-volume. And you're probably sick of doing it yourself.

Once that works, add the next layer. Then the next.

Ninety days from now, you could have a delivery stack running your program while you focus on growth.

---

Find out where your IP stands

Not sure how ready your IP is for this kind of system? That's exactly what our assessment is built for.

It takes five minutes. You'll get a clear picture of how monetisable your IP is right now and what's blocking you from scaling delivery.

Take the IP Monetisation Assessment and see where you stand.