TL;DR: Managing multiple clients with AI sounds great until every session turns into another briefing, one big Project mixes up accounts, and your client skills pile into a mess. So today we’re fixing that. You’ll build one AI system that stores your shared processes once and gives each client their own context, so Claude or ChatGPT delivers client-ready work without a fresh briefing every time. Set it up once, scale it to every client you take on.
Last month, I consulted three of my founding subscribers who all run marketing agencies. All three use AI in their ongoing client work, and the same questions came up in every conversation:
How do you scale AI-powered work across multiple clients?
How do you organize the context shared across your work and the context specific to each client?
Do you need separate skills for every client and every process or asset you create for them?
These are the questions we are going to answer today as we build an organized AI system for your client work.
Because as AI becomes a bigger part of your ongoing work, a few problems start to appear:
Without a saved AI system for each client, every work session starts with another briefing. You have to give your AI assistant the right account context before the work can begin.
ChatGPT or Claude Projects solve this poorly. One large Project can mix information from different accounts. One Project per client creates a long list to manage, with no clear place for the methods shared across your work.
Client-specific skills create another problem. Once you add separate skills for every client’s writing, visuals, and other work alongside your own, your Claude or ChatGPT account can quickly fill with dozens of skills and become a mess.
Creating a dedicated AI system for every client takes time. You have to organize the materials, create the context, and build everything the AI needs for each account. Across a full client list, this can take weeks.
Shared methods and client information need different homes. Some processes apply across all client work, while other instructions, assets, and decisions belong to one client. Storing them together creates duplication and makes it easier for context to get mixed.
This article will help you solve all of them.
And while I’ll use a content and branding agency as the main example throughout this article, the same system works when you manage multiple brands, products, or areas of work.
So today we’ll build on the local folder structure from my Cowork operating system. Although I originally built it in Cowork, you can use the same structure with ChatGPT Work and other agentic setups too.
My goal is for you to have an initial working version today: your client work organized, your reusable processes ready to scale across clients, and a system you can keep expanding as your agency grows.
I’ll give you the structure, the complete prompts, and the exact steps so Claude or ChatGPT can do most of the setup for you.
Here’s what we’ll cover:
We’re going to build this in the same order you would build it for your own agency:
Step 1: Choose the AI skills and workflows your client work needs
Step 2: Build your AI operating system (the foundation)
Step 3: Import your existing client materials
Step 4: Build each client's AI system with an orchestrator
Step 5: Test the system across multiple clients
Step 6: Scale the system to the rest of your agency
Your AI client system is ready
Every step includes a grey Do this now box. Follow those boxes in order if you want to build as you read.
The explanations around each step show what Claude is doing and why each part exists, so you also understand how the system works once it’s running.
Step 1: Choose the AI skills and workflows your client work needs
If you run a marketing agency, that may include repeatable systems for writing, branding, design, social content, ads, reports, research, or websites.
For example, if you write copy for clients, you might have one copywriting skill based on your process. Later, you’ll adapt it to each client’s brand, voice, audience, positioning, and how they want to come across.
If you need a starting point, I’ve created several skills you can use:
My AI writing system includes a Voice DNA skill that teaches AI how you or your client writes. The Audience Profile skill keeps the target audience in context, while the Business Profile skill gives AI the business context and objectives behind the work.
A Draft to Final skill learns how you or your client think and revise from a few finished pieces and their drafts, so AI drafts come back close to what you’d approve.
A brand skill applies the client’s visual identity across documents, decks, websites, and other assets. It can also become the foundation for other visual skills.
A carousel Skill creates branded carousels for you or your clients and can be adapted for individual social visuals.
My image-generation Amplifiers create images and infographics inside Claude.
If you run another kind of agency, your skills will be different. The principle remains the same: identify what repeats across clients and what needs to change for each one.
Do this now
Gather them, at least mentally. You don’t need to go on a scavenger hunt through every folder and organize everything yet. Just know which skills, workflows, frameworks, templates, or checklists you already have and want to use across clients.
From this point on, I’ll assume you have at least one skill or repeatable workflow. It can be mine, adapted from mine, or your own. Later, we’ll customize it for each client and organize the separate versions inside your OS (automatically).
Step 2: Build your AI operating system (the foundation)
Those foundational processes we covered in Step 1 need an organized place to live. That’s where an operating system comes in.
I covered this in a previous article, but I’m repeating the essentials here because building this OS really is the biggest upgrade you can make to how you work with AI (not only if you run an agency).
How the AI operating system is organized
“AI operating system” sounds fancy, but it starts with something we’ve all used to organize our work since computers became a thing: folders.
Long story short, the AI OS is one folder on your computer, synced through Google Drive so it can also run in the cloud when your computer is closed.
Inside it, your work, context, skills, and processes are organized in a specific way. The files tell your AI where to look, what to use, what belongs to each client, and what it should remember. That’s what makes it a smart operating system and not just a folder.
If you missed my first article, think of your AI OS like a company with three levels:
The root is company management. This is the main folder for your entire business. Your AI reads its instruction file (
CLAUDE.mdorAGENTS.md) first to understand how the system is organized and where to go for different kinds of work. Its memory files (MEMORY.mdandARCHIVE.md) keep decisions that apply across the company, while its Resources folder holds general context about you, your business, and your goals. If you ask for client content, it sends your AI to Clients. If you ask about an invoice, it sends it to Finance.Workspaces are departments. Clients, Marketing, Finance, Sales, or Team Management can each have their own workspace, with instructions, memory, and resources for that area. The Clients workspace’s Resources folder also holds the processes you use across several clients. For example, the process you use to write LinkedIn hooks lives here once instead of being duplicated across every client folder.
Projects are individual clients. Inside Clients, Client A gets one project folder, Client B gets another, and so on. Each project has its own instructions and memory, along with a Resources folder specific to that client (business context, audience, voice, brand, approved claims, current work, references, and assets). When you ask for something for Client A, your AI combines the shared process from Clients with only Client A’s information.
And because the folder is synced through Google Drive, your team can work from the same company context and processes, while you can schedule tasks to run inside it even when your laptop is closed.
How your AI client system works in practice
Now let’s see what happens when you ask your AI to create something for one client:
By moving through these levels and reading the files named in the instructions, your AI gets all the context it needs for that specific task and client. You don’t have to brief it again every time you do client work.
Build your AI OS with two Amplifiers
Now that you understand how the system works, it’s time to build yours.
We’ll use two Amplifiers for this. The Cowork Workspace Planner helps you decide how the OS should be organized and creates the blueprint. The Cowork Workspace Builder takes that blueprint and builds the complete folder structure and files for you.
And don’t worry about remembering every file or deciding where everything goes. The Amplifiers will do the building. You’ll provide the context and follow their guidance.
Do this now
Open Cowork and run the Cowork Workspace Planner Amplifier.
Paste this into a new session:
Use the Cowork Workspace Planner Amplifier and help me brainstorm my AI OS.
Tell it about your business, the areas of work you want inside the OS, and the skills and processes you identified in Step 1. Make sure Clients is one of the first workspaces you want to build.
If you already have an OS without a Clients workspace, tell the Planner you want to extend it instead of starting again.
The Planner will ask about your work, clients, team, existing skills, repeatable processes, tools, and preferences. It will help you decide what belongs at the root, workspace, and project levels, then create the complete blueprint.
Copy the entire blueprint and open a new Cowork session.
Paste it, then write:
Use the Cowork Workspace Builder Amplifier. Here’s my blueprint:
The Builder will create the folders and files, save your repeatable processes as workflows, connect your existing skills and tools, build the first workspaces and projects from your blueprint, and guide you through syncing the local folder with Google Drive.
When it finishes, you’ll have the first working version of your AI OS.
Next, we’ll add your existing client materials.
Step 3: Import your existing client materials
At this point, the structure exists. Now Claude needs the source material for each client.
You probably already have most of it somewhere: brand guidelines, meeting notes, writing samples, old campaigns, proposals, research, visual assets, or anything the client sent you.