AI Orchestration

How we run AI across every brand and client. The full stack, nothing held back.

We don't use AI.
We run an AI Operating System.

Most teams bolt a chatbot onto the side of the business. We built a full stack underneath it: frameworks, skills, memory and connectors that do real work and get sharper every week. This is the whole thing.

Models

The brains

Frameworks

A business in a box

Skills + agents

The muscle

Memory

The mind

Connectors

The hands

Automations

The heartbeat

00 · Fundamentals

Fundamentals: Get Set Up

The once-off plumbing. Subscribe to Claude, install the apps and CLIs, set up git, before you build anything.

Before any of this, get the plumbing right. This is the once-off setup: the model, the apps, the tools, and version control. Nothing clever, just done properly so nothing trips you up later. Skip it and every stage after this fights you.

01

Subscribe to Claude

Get a Claude paid plan. This is the engine. Everything else sits on top of it.

02

Install the desktop app

Download Claude Desktop. It is where the everyday, in-your-tools work happens.

03

Install the CLIs

Claude Code (the driver), Codex (the reviewer), and Antigravity for Gemini (the eyes and ears). Sign in to each.

04

Set up git + GitHub

Install git, sign in to the GitHub CLI, and put your work in a private repo. Version control is the safety net under everything.

05

Add md-architect

Install md-architect from our GitHub. It builds the context files (CLAUDE.md, AGENTS.md) the next stage depends on.

The interactive sign-ins are yours to do. Claude will hand you the exact commands and wait. Never paste a password or key into the chat. Secrets live in a gitignored .env file.

01 · The big idea

An Operating System, Not A Chatbot

Most teams bolt a chatbot onto the side of the business. We built a full stack underneath it.

Think of how a phone works. The apps you see sit on top of an operating system you don't. Our setup is the same. Every job runs on one shared stack, so nothing starts from scratch and nothing gets lost.

How the system is structured

The AI Orchestrator

Runs the whole stack, one accountable brain

Internal team

Runs our own shop

Clients

Deployed per engagement

Agents

Focused workers, one job each

Skills

Packaged methods they run

The models, the main interface

Claude

Gemini

Codex

Dashboards

One place to see the whole thing

Memory systems

The wikis + Pinecone

Sit alongside every layer and link them together, so nothing starts from scratch.

The headline

Most AI setups reset every time you open them. Ours compounds. Every meeting, sales call, and work session feeds the stack, so the next answer starts from what we already know.

02 · Where work happens

One Driver, Two Specialist Brains

Claude runs the show. Two other top models check its work, so we catch what one model alone would miss.

We run on Claude as the main driver. Then we wire in two other top models for the jobs they do best. They check each other's work, so we catch mistakes one model alone would miss.

Claude

The driver. Runs every session, build, and decision. Stays in front of you.

Codex

The reviewer. A different model with different blind spots. Checks the work for bugs and risk.

Gemini

The eyes and ears. Reads video, audio, and PDFs, and scans a whole codebase in one pass.

The rule that matters most

No model reviews its own work. Same model, same blind spots. So every important piece of work gets a second pass from a different architecture before it counts as done.

Three ways to work with it

People think every AI tool is the same chatbot. It is not. There are three modes, as different as watching a cooking show, hiring a chef, and running a restaurant. We pick the mode that matches the job.

Cooking show

Chat

You ask, it answers, you still do the work. Great for ideas. This is where most people stop.

The chef

Co-work

An assistant working inside your tools: email, calendar, docs, finance. It does the task with you.

The restaurant

Code

Runs the whole operation. Builds, automations, the full stack. This is where we live.

The simplest way to picture the stack

Think of a car. The model is the engine. The harness is the chassis around it. An agent is a self-driving car. Knowing which part is which is half the battle.

The engine

= The model

Claude, GPT, Gemini. Raw intelligence. You swap it as better ones arrive, without rebuilding the car.

The chassis

= The harness

The software around the model: the app, the connectors, the controls. What turns an engine into something you can drive.

Self-driving

= The agent

You give it the destination and let go. Most people do not need a self-driving car yet. Most do not need an agent yet either.

03 · Frameworks

Fulfilment Frameworks, Deployed Per Client

A complete way of running one type of work, built once and dropped into each client project.

These are our fulfilment frameworks: complete, packaged ways of running one type of work that we deploy into each client. We build the master once, strip out anything private, and deploy a fresh copy per client. Built once, deployed again and again. The examples below are ours, by category.

Master framework→
Strip private data→
Deploy per client
Revenue

Revenue framework

A step-by-step playbook for turning an AI engagement into measurable revenue.

Consulting

Consulting framework

A build that gives a client their own AI system, made from their own know-how.

Dev

Build framework

Quality gates and test-first builds for shipping software cleanly.

Growth

Growth framework

Sales, content, and offer design for personal-brand growth.

Content

Content framework

Content and search, so the work gets found by people and by AI.

Assistant

Assistant framework

A persistent AI assistant that runs in the background and spawns specialists. Custom-built, not an open-core agent off the shelf.

The shape every framework shares

Orchestrator agent→
Sub-agents→
Skills + scripts

Whatever the framework does, the setup is the same. One orchestrator agent runs the job. It hands focused work to sub-agents underneath it. Each sub-agent is backed by skills and scripts that carry the actual method. Same shape, different job.

Enjoying the guide?

Enter your details to unlock the remaining insights. We'll also send you the complete guide as a reference.

No spam. Unsubscribe anytime.

04 · The internal team

The AI Staff That Run Our Own Shop

We don't just build AI for clients. We run our own business on it. This is how we know it works.

We don't just build AI for clients. We run our own business on it. A standing team of AI staff handles the day-to-day, the same way a company runs on people. This is how we know the stack works. We live in it.

The Orchestrator

Owns the master view. Tracks what is overdue, keeps priorities straight.

EA

Inbox and calendar. Scheduling, briefs, and meeting prep.

Sales Team

Pipeline and follow-ups. CRM kept current. Lead, then specialists.

Marketing Team

Brand voice and audience. Strategy, ads, and distribution.

Content Engine

Turns briefs into assets. Posts, emails, scripts. Always on-brand.

AI Stack

Where we log how the whole system runs, and vet any new piece before it goes in. The control room for the operating system.

Drafts only, human approves

AI is an enthusiastic intern, not a magic button. So we run 10/80/10. 10% planning, 80% the AI doing, 10% a human making it right. The two 10%s are where the judgement lives, and where the slop gets killed. Nothing goes out without a human yes.

05 · Skills + agents

The Muscle: Skills That Fire On Cue

Packaged know-how that sits quiet until you say the right thing. How to make them, find them, and vet them.

A skill is a packaged piece of know-how. It sits quiet until you say the right thing, then it fires with a full method behind it. An agent is a focused worker we hand a single job. We have thousands of skills ready across the stack.

You ask→
Trigger spotted→
Skill fires→
Checked, then delivered

A few we lean on

Deep research for evidence-led, cited answers. Voice skills that keep every draft on-brand. Accountability and planning for the weekly rhythm. Tribunal for pressure-testing big decisions from five angles.

Right-sized cost

Every job is routed to the cheapest model that can still do it well. Heavy thinking gets the top model. Simple tasks get a fast, light one. We are not paying premium rates for simple work.

Make, find, vet

Make one

Found yourself explaining the same process twice? Write it down once as a skill: the trigger that fires it, the steps, the standard. Now it runs the same way every time, for anyone.

Or find one

You rarely need to start from scratch. Grab a proven skill and adapt it. We keep a set of public, ready-to-use skills you can lift straight off our GitHub.

Always vet it

Never drop a new skill straight into a live account. Run it through your AI stack first, so it proves it adds value, and let the stack improve it using the context it already knows about your business.

Grab our public skills

We keep a set of ready-to-use skills open on GitHub. Take what is useful, run it through your own stack to fit it to your business, and go. github.com/ai-orchestrators

Why this beats a generic AI assistant

A generic assistant improvises every time. A skill carries a tested method, so the output is consistent no matter who runs it or when. That is the difference between "the AI had a go" and "the AI ran our process."

06 · Memory

The Wikis That Compound

An AI model does not get smarter on its own. What compounds is the memory you give it. So we built the wikis.

Here is the thing nobody tells you: an AI model does not get smarter on its own. The weights are frozen, the same every session. The self-improving agent pitch is marketing. What actually compounds is the memory and context you give it. Structure beats raw intelligence. So we built the structure.

That structure is a set of wikis: living, interlinked maps of how the business connects. Each one is plain text the AI reads before it acts. The more we work, the sharper the maps get, even though the model underneath never changes.

The wikis we run

The Meeting Wiki

Every call and meeting, synthesised and filed. Ask what we decided with a client, or pull a quote, and it comes from the wiki, not a frantic transcript search.

The Content Wiki

Brand voice, positioning, and every asset we have shipped, mapped so new content starts from what already works instead of a blank page.

The Stories Wiki

The real anecdotes, metaphors, and proofs in the founder's own words, so any piece of content sounds like a human, not a generic AI draft.

Compounding, not resetting. Each session adds to the maps, and each client's wiki is walled off in its own space. The AI gets sharper every week, and the truth lives in plain text that survives any rebuild.

Want to build one for your calls? The Meeting Wiki Brain guide is the step-by-step, with a build-it mega prompt, for coaches, consultants and course creators.

07 · Redundancy

Pinecone: The Backup Brain

The wikis are the mind. Pinecone mirrors everything into a portable store, so you never lose your context.

The wikis are the mind. Pinecone is the redundancy. It is a vector database that mirrors everything we know into one searchable store, so the knowledge can be called on any time, from any tool, even as the rest of the stack changes around it.

Search by meaning

Everything gets indexed in a vector database. Ask in plain words and it finds the right knowledge by meaning, not keyword matching. The AI pulls the exact insight it needs, fast.

A redundancy layer

It quietly collects all of it in one place. If a tool changes or a wiki moves, the knowledge is still there, captured and searchable. A safety net under the whole system.

Portable, not locked in

Because it is a standard vector store, you can call on it from anywhere and plug it into other solutions. No single tool owns your memory, so no single tool can hold you hostage.

Reduces dependency

The wikis are the source of truth. Pinecone is the backup brain that means you are never one outage or one price hike away from losing your context.

Two layers, one job: never lose what you have learned. The wikis keep it structured and human-readable. Pinecone keeps a portable copy ready for whatever comes next.

08 · Connectors

The Hands: Live Links Into Real Tools

Knowing things is no use if you can't act. Dozens of live links into the tools a business already runs on.

Knowing things is no use if you can't act. We have dozens of live connectors into the tools a business already runs on, so the AI can actually do the work, not just talk about it.

Comms

  • Email
  • Slack + Teams
  • Drafts + threads

Calendar + PM

  • Calendars
  • Task + project boards
  • Time blocking

Finance

  • Accounting
  • Invoices + reports

CRM

  • Contacts
  • Pipelines
  • Per-brand setups

Build + deploy

  • Databases
  • Hosting
  • Code repos

Research + data

  • Web crawl
  • Search + SEO data
  • Live docs

Creative

  • Design
  • Image + video

Browser

  • Real web actions
  • Forms + checks

That is the difference between an assistant that suggests an email and one that drafts it in the right inbox, books the meeting, updates the CRM, and logs the invoice.

09 · Build + ship

Dashboards, Pages, And Live Sites

We don't just answer questions. We build and ship real things.

We don't just answer questions. We build and ship real things: live dashboards, interactive pages like this one, and full client websites and apps.

Build→
Preview→
Deploy

Live dashboards

We run dashboards that show the whole operation at a glance. They share one design system, so everything looks like one brand.

Interactive pages

One-off pages built on demand, on-brand, with no template. This page is one of them, built start to finish on our own design system.

Client sites + apps

Full websites and web apps shipped to live hosting with a database behind them when needed.

10 · Automations

The Heartbeat: It Runs Itself

A lot happens with nobody at the keyboard. Jobs that run on their own clock, every day.

A lot happens with nobody at the keyboard. Dozens of jobs run on their own clock, every day. They keep the stack fresh, file new knowledge, and flag anything that needs a human.

Set time arrives→
Job runs on its own→
Writes back to memory

What runs on a clock

A daily stack refresh. A monthly cost audit. Meeting notes filed automatically. Content and search reports. Sales follow-up sweeps. All without a prompt.

The system watches itself

A daily guard checks every part of the stack for drift and flags problems early. The operation keeps its own house in order.

11 · Security + trust

Run AI At Scale, Lock It Down

The moment you run AI across a team, your AI accounts become a target. We learned this the hard way.

The moment you run AI across a team, your AI accounts become a target, the same as a bank login. We learned this the hard way once. Now we treat security as part of the cost of running AI, not an afterthought.

Accounts locked down

Individual logins, not shared magic links. 2FA everywhere, authenticator over SMS. Hard spend caps so a breach can't run up a bill.

Every skill vetted

A third-party skill can quietly steal data. So nothing runs until a trusted AI agent has checked it first. We never install unvetted tools into a live account.

Client data walled off

Each client's knowledge is locked to its own space and physically cannot reach another. Built into the engine, tested against attacks, not left to trust.

Human on the trigger

Nothing is sent, posted, or paid without a person saying yes. The AI prepares the work. A human pulls the trigger on anything that leaves the building.

12 · Making it land

The Build Is The Easy Part

Building the AI is no longer the hard part. Getting people to use it is. So we treat adoption as a deliverable.

Here is the honest bit most agencies skip. Building the AI is no longer the hard part. Getting people to use it is. So we treat adoption as a deliverable, not a hope. And we never force AI where it does not belong.

Champion→
Workshop→
Hands-on training→
Docs + support

Adoption is the real blocker

The tech work is the easy 20%. Resistance is the other 80%. We lead with change management so the build actually gets used, instead of sitting on a shelf.

Orchestrate, don't build

You don't have to be the best builder to win. The one who stitches the right tools together and owns the outcome captures the value. We sell the result, not the tech.

Not everything needs AI

Sometimes the right tool is a plain, deterministic automation with zero AI. Forcing AI into a process that doesn't need it adds risk for no gain. We pick the tool that fits.

Interfaces→
Frameworks→
Skills + agents→
Connectors do the work→
Memory feeds it back

The payoff

Most teams get an assistant that re-learns the business every single session. We built one that gets sharper every week the work gets done. That is the whole point of running AI as an operating system instead of a one-off tool. The AI Orchestrator OS is where we build this with you, on your own IP.

This guide is the spine of the AI Operating System handbook. Once several teams are running, the cockpit that sits on top of all of this is the AI command centre.

Built and run by DevWiz, The AI Orchestrators, and Njin.