You were one of the small group of agencies who tried the shared team brain before it had a name, a proper app, or half the things it has now. That experiment grew up. It's called Agency Brain, it's live, and this page walks you through what changed and the short path to move your team onto it.
Right now everyone on your team runs their own AI in their own little world. Each person builds up their own context, their own prompts, their own way of doing things, and none of it reaches anyone else. When someone works out a brilliant way to build an audit or write a report, that stays locked on their laptop.
Agency Brain is one shared brain your whole team works from. A skill a builder makes today, everyone can use tomorrow. The context you write once, every session on the team reads. That's the whole point: stop everyone re-solving the same problems alone.
The beta proved the idea worked. Agency Brain is that same idea, finished: a proper app instead of git commands, three clear roles instead of "head scout and everyone else", and team members who never have to touch anything technical.
/save and a few git steps by hand| Team Brain (beta) | Agency Brain (now) | |
|---|---|---|
| Staying in sync | You ran /save and git steps | A background app pushes and pulls for everyone automatically |
| Setup | Long checklist on a website | The app's wizard walks you through it once |
| Roles | Head scout / team member | Owner / Scout / Team |
| Team and GitHub | Everyone needed GitHub | Team members never touch GitHub |
| Managing people | On a dashboard website | Inside the app, in the Command Centre |
| The template | Older members template | Cleaner, agency-tuned skill set |
You don't drag your old beta setup across. Everyone installs a clean copy of the new template so the whole team stands on the same, consistent foundation. It also skips the rough edges that come from carrying an old beta forward. The agencies who've already moved across did exactly this, and it was quick and clean. Nothing you built is lost. See step 4 on the "What you do next" tab for bringing your custom skills across.
The old "head scout and everyone else" split becomes three clear roles. The big idea: the people who build the skills and the people who use them are different jobs, and they get different tools and different access, so nobody breaks anything by accident.
You're the principal. In the beta you were the "head scout"; now you're the Owner. You hold the brain, add and manage everyone, and decide who builds and who uses. You're a full Ads to AI member, so Agency Brain is already included for you.
What only the Owner can do: host the brain on the agency's own GitHub, add and remove people, and keep truly private docs (finances, sensitive notes) in a separate private repo that only you can see. You don't have to be the technical one, you can hand the building to a Scout, but the brain belongs to you.
A Scout builds and improves the skills the rest of the team uses. This is the technical seat. It might be you, or a hands-on teammate (plenty of agencies keep the Owner and the first Scout as two different people). Scouts are full Ads to AI members, and a Scout is the only role that costs extra.
Why a different tool: Scouts use Claude Code (run inside the free Cursor editor) for full power and speed. They can see and change the .claude folder, where the skills, agents and hooks live. That's the workshop.
You need at least one Scout (the person who builds), and often the Owner is also the first Scout. Everyone else can be Team.
Team members use the skills the Scouts built, day to day. They don't build, they do the work: run the audit, draft the report, pull the numbers. This is most of your people, and it's free up to your plan's cap.
The safe-by-design part: Team members can read and run every skill, but they can't change the .claude folder, so nobody accidentally breaks a skill everyone relies on. They still have full write access to the working folders (clients, data, projects).
Nothing technical, ever: a Team member installs one app, signs in with their email, and they're in. No GitHub account, no terminal, no setup. They use the friendly Cowork app, not Claude Code.
A person's AI never gets more access than the person. If a teammate can't see a file, neither can their Claude session. Each person works from their own machine, with their own permissions. That's why an Owner can keep private docs private and a Team member can't wreck the skills, all at the same time.
In the beta, keeping in sync was a job you had to remember to do. Now a small app does it for you. Here's the whole thing in one picture.
All the files (skills, context, client folders, notes) live in a single private GitHub repo that your agency owns and hosts. That's the source of truth. GitHub is used because it's fast and it tracks every version, which matters a lot for AI work. Nobody edits it on the website; they just work in their own copy.
Each person installs the Agency Brain app. It sits in the menu bar on a Mac, or the system tray on Windows, and runs in the background. It watches the brain folder on that machine, sends changes up, and brings everyone else's changes down. No /save, no git, no thinking about it.
This is the big beta fix. A Team member does not need a GitHub account. When they sign in, the app gets them a short-lived key behind the scenes that only works on your one repo. They get the latest skills and context without ever seeing GitHub.
The brain carries CLAUDE.md files. The one at the top loads into every session automatically, and the one inside a folder loads whenever someone works in that folder. So the right background is always in front of the AI, without anyone pasting it in.
An agency runs on valuable access: your Google Ads API login, and often keys for other tools too. You really don't want a copy of those sitting on every team member's laptop. Agency Brain's answer is a secure vault, so the whole team can use that access without anyone ever holding the actual keys.
To let someone's AI pull live Google Ads data the old way, the credentials have to sit on their machine. Multiply that across a team and you've got copies of a powerful key in a lot of places. Lose a laptop, or have someone leave, and you're rotating credentials and hoping you caught every copy. And a raw key on a machine can accidentally end up pasted into a chat or a log.
You set up one small secure gateway (a Cloudflare worker, your Scout handles it). The real credentials live there, encrypted, in a single place. Each team member gets just one thing: a personal token you can switch off any time. When their AI needs data, it asks the vault, and only the answer comes back.
Someone leaves, you revoke their token. Nothing else changes, no credential to rotate, nobody else is disrupted.
A lost or stolen machine has no keys to steal, just a token you can kill in seconds.
This is the point for Cowork. Your team uses the simple Cowork app and isn't technical. The vault lets them pull live Google Ads data safely without ever handling a credential.
Because the secret never reaches the session, it can't accidentally end up in a message or a log.
The same trick works for other keys your team leans on (image generation, the AI models themselves, and more). The direction is one secure front door for your agency's keys, so the only secret on anyone's machine is a single token you control. Google Ads is the one that's ready today; the rest is where this is heading.
Your Scout deploys the vault once and hands each person a token. It's a guided setup, not a self-serve button yet, so it's something I'll walk you or your Scout through when you're ready. Worth knowing it's there, because it's a big part of why a shared agency brain is safer than everyone running their own.
This is what you, as the Owner, do to move your team over. Tick things off as you go, this page remembers your progress on this device. Your teammates have it even easier: they only do the last step.
This is the one thing that trips people up on day one. On Windows, grab it from git-scm.com and pick the middle option when it asks about PATH, then fully quit and reopen. On a Mac it's usually already there. Team members can skip this.
Go to m.ads2ai.com/agency-brain/scout-path. It's the step-by-step for standing your agency brain up, and it walks you through everything on this list with screenshots. Start there and keep it open.
Download the Agency Brain app from ads2ai.com/downloads (it auto-detects your Mac or Windows). Open it and sign in with your email, you'll get a 6-digit code. Both installers are properly signed now, so no scary warnings to work around.
The app's wizard does the one-time GitHub bit for you: it creates your fresh agency brain from the template and connects it. Use a GitHub organisation your business owns, not a personal account, so the brain survives someone leaving. Then pick a folder and the app clones the brain onto your machine and starts syncing. Optional: if you've built custom skills or client context in your old brain, use the ready-made migration prompt further down this page (just below) to copy the right bits across. Your old brain is left untouched.
Pick who builds. It can be you, or a hands-on teammate. Then, inside the app, open the Command Centre, click Add a person, give a name and email, and choose the role. Scouts (builders) use a paid seat and switch on automatically a few minutes later. Team members are instant and free. No emailing me lists, you run it yourself.
Tell each teammate: install the app from ads2ai.com/downloads, sign in with the email you invited them under, pick a folder, done. They're now in sync with everyone, with the friendly Cowork app for their day-to-day. That's the entire team-member setup.
Your fresh agency brain starts clean, on purpose. To pull your custom skills and client context out of your old Team Brain, paste one of these into Claude Code while you're in your new agency brain folder. Both leave your old brain completely untouched. Change the folder path on the first line to your real one.
I'm moving from my old Team Brain into this fresh Agency Brain, and I want to bring
the right things across without breaking anything.
My old brain is at: ~/Projects/OLD-BRAIN-FOLDER (change this to your real path)
Treat the old brain as READ-ONLY. Never edit, move, or delete anything inside it.
We are only ever COPYING out of it.
Work in stages and STOP for my approval before copying anything:
STAGE 1 - Understand what's there.
- Read the CLAUDE.md at the root of my old brain, plus any folder-level CLAUDE.md files.
- List the top-level folders and the notable files in each.
STAGE 2 - Recommend what moves, and where. Put every folder into one of:
(a) BRING ACROSS to this shared agency brain:
- custom skills I built that are NOT already in this agency template
(compare against .claude/skills/ here first, and skip anything that already exists)
- client / customer / account context, call notes, reports
- processes, SOPs, checklists, report templates
(b) MY PERSONAL FOLDER here (my own notes, my todos, my preferences)
(c) LEAVE in the old brain only:
- anything private or sensitive (finances, personal projects)
- research / newsletter / youtube folders (the agency brain handles research separately)
STAGE 3 - Show me the plan as a simple table (source -> destination -> why) and WAIT
for me to approve or edit it. Do not copy anything yet.
STAGE 4 - Once I approve, copy the approved items into the right places here, without
overwriting anything that already exists.
Hard rules, never break these:
- NEVER copy .env files, credentials.json, token.json, google-ads.yaml, service-account
files, or anything inside a secrets folder. If you spot one, tell me, do not copy it.
- NEVER copy the root CLAUDE.md or context/business (this agency brain has its own).
- NEVER overwrite an existing skill here. If names clash, show me both and let me choose.
- When you're done, give me a short list of what came across and what you deliberately skipped.
Help me give this agency brain the core context about my agency, so every session starts already knowing who we are. Interview me a few questions at a time (not all at once) to capture: - The agency: our name, what we do, the kind of clients we work with, our tone of voice. - The team: who's on it, their roles, and who does what. - How we work: our main tools, our reporting cadence, our do's and don'ts. - Our clients: a short profile for each active client (their goals, accounts, quirks). As we go, write what I tell you into sensible files here: a top-level context file for the agency, and one folder per client under clients/. Keep the writing tight. Do not invent anything I haven't told you. When we're done, show me the folder tree you created.
We're still in beta, so if anything snags, I post fix-it prompts on the install page: m.ads2ai.com/agency-brain/install. And just reply to my email any time. I'd much rather hear "I'm stuck on step 4" than have you give up on it. I'm keeping an eye on each of you to make sure you land it.
No. Moving onto Agency Brain is included in your Ads to AI membership. You get the app and free Team seats up to your plan's cap. The only paid part is Scout seats, the extra builders, and only if you want more than one person building skills. If you're the only builder, there's nothing new to pay.
No, nothing gets deleted. This is a clean install of the new template, sitting alongside what you have. Your old brain stays exactly where it is, and you can copy across anything custom you built using the migration prompt (step 4). The old beta dashboard just stops being needed, because everything now lives inside the app.
A clean copy of the agency template means your whole team stands on the same, consistent foundation, with the newer agency-tuned skills. Carrying the old beta setup forward drags its rough edges along with it. The agencies who've already moved across did exactly this, a fresh install, and it was quick and clean. You keep your custom work by copying it in, not by upgrading in place.
No. This was the biggest friction in the beta and it's gone. A Team member installs the app, signs in with their email, and they're in. The app handles all the syncing behind the scenes with a key that only works on your one repo. They never see GitHub. Only builders (Scouts) work near that layer, and even then the app does the syncing for them too.
Scout if the person builds and improves skills, is comfortable being a bit technical, and will use Claude Code in Cursor. It's a paid seat. Team if the person just uses the skills to do their work, and wants the simple Cowork app. It's free. Most of your people are Team. You need at least one Scout, and it's often the Owner or one hands-on person to start.
That was the beta dashboard. Everything it did, adding people, managing the brain, getting help, now happens inside the app's Command Centre. You won't need the old website. Even the agency setup site is basically just a download link now; the app runs the whole setup itself.
Yes, but it's faster now, so don't let that put you off. There's no long checklist to grind through anymore. The app's wizard does the GitHub setup for you, and a Team member's entire job is "install the app and sign in." Whatever got you stuck before is very likely handled for you this time.
Scouts use Claude Code, run inside the free Cursor editor, for full power, speed, and a visual view of the files they're changing. Team members use Claude Cowork, the simpler app, to run the skills without touching how they're built. Same shared brain underneath, two different front doors depending on whether you're building or doing.