Free playbook · for agency owners

Every question about a client still routes back to you.

Build a working company brain for one client, in one afternoon, using the ChatGPT or Claude account you already pay for. No new tools. No code. And an honest account of exactly where it stops working.

Company email addresses only. One email, PDF attached.

Cover of The One-Client Brain, a build playbook for agency owners by EfficiaLabs.

The problem

Fourteen tools, one index, and the index is a person.

You hired people. You wrote SOPs. You keep a ChatGPT tab open all day. And everything still waits on you.

Why the obvious fixes failed

None of them failed because the tool was bad.

They failed because the context was scattered, and the only complete index of it was you.

  • You dumped it all in Drive

    And drowned in nested folders. You still can't find the SOP you wrote last month.

  • You wrote a proper SOP library

    By month four half of it was outdated. Nobody went back to update it.

  • You keep AI tabs open

    And re-introduce yourself fifty times a day. Brand voice, brief, SOP, again.

  • You hired more people

    The team got bigger. The decision queue stayed exactly the same size.

The part everyone gets wrong

You've been documenting the wrong half.

Most people document process. That's the easy half, and the half that rots fastest. The hard half, the half your team actually pings you about, is judgment. Knowing which client request to push back on and which to just do doesn't transfer by writing an SOP.

  • Process

    “Export the report on the 1st.”

    Transfers fine in an SOP. Also the part that goes stale first.

  • Judgment

    “Push back on this one, absorb that one.”

    Never transfers in an SOP. This is 90% of what they ask you.

What's inside · step one

Five files. About ninety minutes.

Everything the brain knows comes from these. Write them in Docs, Notion, or plain text. Reasons matter more than decisions: a decision tells your team what you did once, a reason lets them make the next call without you.

  • 01

    01-client.md

    Who they are, and what they're really buying.

  • 02

    02-promise.md

    Signed scope, plus every change since, dated.

  • 03

    03-voice.md

    How they sound. Approved work, banned words.

  • 04

    04-how.md

    The 2–3 SOPs that actually apply here.

  • 05

    05-decisions.md

    Every judgment call, with the reason attached.

Page 4 of the playbook: the five files to write, each with what belongs in it.
Page 5 of the playbook: worked examples of a decision log, each entry paired with the reason behind it.

The honest part

By week eight, it will lie to you.

Here's what nobody selling you an AI workflow mentions. Project knowledge is a snapshot, not a connection. You upload five files and the tool takes a copy. You edit the original; the copy never finds out. It won't announce this. It'll answer with total confidence from a scope document that changed a month ago, and cite its source, which makes it more convincing, not less.

There are two full pages on this in the playbook, plus the manual fix that genuinely works. Build the thing first, then read them.

What Claude and ChatGPT Projects can't do

None of this makes them bad tools. I use both every day. It's just what you're working with.

CapabilityProjectsWhat that means in practice
Stays current with your files✕Knowledge is a copy taken at upload. Edit the original and the brain never finds out.
Reads Drive, Slack, email, CRM~Connectors fetch live when you ask, but nothing flows in on its own. Someone has to open a tab.
Holds your whole agency✕One client fits comfortably. Twenty-eight don't.
Knows when something is out of date✕No concept of freshness at all. The playbook gives you a hand-built substitute.
Reviews work before a client sees it✕There are no triggers. The check that matters most is the one most likely to get skipped.
Answers accurately from what you gave it✓This part genuinely works, and it's why the afternoon is worth it.
Page 11 of the playbook: a table of what Claude and ChatGPT Projects cannot do, with what each limit means in practice.

By tonight

What you'll have when you finish

  • A working brain

    A project in Claude or ChatGPT holding everything about one client.

  • Sourced answers

    Every answer cites which file it came from, so anyone on your team can check it.

  • A decision log

    The thing that actually transfers your judgment, not just your process.

  • Copy-ready instructions

    The full project instruction block. Paste it in, change the client name, done.

  • The ten-minute test

    Five real questions to throw at it, so you know within a day whether it works.

  • The honest limits

    Two pages on where it breaks, when, and the fix that holds it together.

Page 8 of the playbook: the full project instruction block, ready to paste into Claude or ChatGPT.

Who wrote it

Vaibhav Sharan, co-founder of EfficiaLabs.

Four years at Accenture building operational systems, now running EfficiaLabs entirely on the thing described in this playbook. I built it for myself first.

Email me if you get stuck building yours. I'll answer either way.

Talk to Vai (co-founder)

Before you go

Not ready to talk? Get your next client deliverable out faster anyway.

Take the 230 AI skills I give agencies for free. They work in Claude, ChatGPT and Gemini, in 10 folders, one for each service you sell.

Use your agency email. I'll send the link there.