One folder per client that your agent can read
How to organise files for AI agents doing client work: one plain folder per client, a short brief file at the top, Markdown notes, and what never goes in.
A florist's website brief lives in an email thread from January. Her brand colours are in a PDF on your desktop. The note that she hates the word "bespoke" is in a chat with your agent from three weeks ago, which the agent no longer remembers. Every time you start work on her site, you spend the first ten minutes pasting the same things back in, and you still forget one.
For a one-person AI agency, the fix is dull and it works: one folder per client, laid out the same way every time, with plain text notes your agent can read on its own.
Here's the short version. Give each client one folder with a plain name. At the top, a short brief file: who the client is, what they bought, what matters to them. Underneath, a few subfolders for brand, content, the website and dated history. Keep the notes in Markdown, plain text with simple marks for headings and lists, so you, the agent and any future tool can open them.
What should a client folder contain?
The same few things for every client, so you and the agent always know where to look:
florist-rose-and-fern/
CLAUDE.md short instructions for the agent
brief.md who they are, what they bought, what matters
brand/ logo files, colours, fonts, tone of voice
content/ page text, menus, prices, photos from the client
site/ the website itself, if you build it here
history/ dated notes: calls, decisions, approvals
| File or folder | What goes in it | Who updates it |
|---|---|---|
brief.md |
The business in a few lines, what they bought, the deadline, words they like and dislike | You, after each call |
brand/ |
Logo files, colour codes, fonts, two or three example sentences in their voice | You, once, then when it changes |
content/ |
The client's own words and photos, as they sent them | You or the agent, filed as they arrive |
site/ |
The website's files | The agent, with your review |
history/ |
One dated note per call, decision or approval | You or the agent, never edited after the fact |
CLAUDE.md |
How the agent should work on this client | You |
Name the folder after the business, not the project. "florist-rose-and-fern" still makes sense when the second project for the same client comes along in a year.
Why keep client notes as plain text?
Because plain text outlives every tool you'll use. Markdown was designed, in the words of its creator John Gruber, as an easy-to-read, easy-to-write plain text format. A heading is a line starting with #. A list is lines starting with -. That's most of it.
Three things follow from that:
- You can read it anywhere. Any text editor on any computer opens a Markdown file. So does a phone.
- Your agent can read and update it without a special connection or export.
- It moves with you. If you change apps, the folder comes with you, whole.
Keep Word documents and PDFs the client sent you in content/ as they arrived, but write your own notes as Markdown beside them. The agent reads a short Markdown summary more reliably than a forty-page PDF.
How does the agent find its way around?
Tell it, in a file it reads every time.
Claude Code's documentation says each session begins fresh, and that CLAUDE.md files, which are Markdown files of instructions, are read at the start of a session. Files in a subfolder are read when Claude works in that subfolder. So a short CLAUDE.md inside each client's folder carries the basics from one session to the next without you pasting anything.
Keep it short. Something like:
# Rose and Fern, florist
Read brief.md before any work.
Brand rules are in brand/voice.md. Never use the word "bespoke".
The client approves every page before it goes live.
Write a dated note in history/ after each work session.
The brief file does the heavy lifting. An AI agent client brief covers what to put in it so the agent doesn't lose the thread halfway through a job. If you use Claude Cowork, the desktop version of Claude for work that isn't code, Claude Cowork explains how it works with folders you choose to share, and the same layout serves both.
How do you set up the folder for a new client?
Copy an empty template folder, so every client starts with the same layout and you never invent a new one at eleven at night. The first setup takes about as long as the kickoff call's notes:
- Copy the template and rename it after the business.
- Write
brief.mdstraight after the kickoff call, while it's fresh: what the business does, who its customers are, what the site must do, the deadline, and any words the owner loves or hates. - File what the client has already sent. Logos and colour codes into
brand/, their words and photos intocontent/, each with the date it arrived in the file name. - Add the first history note: the date, who was on the call and what was agreed.
- Fill in
CLAUDE.mdwith the three or four rules the agent must follow for this client. - Ask the agent to read the folder and summarise it back to you. If its summary is wrong or thin, the brief is missing something. Fix the brief, not the summary.
That last step is the quickest test there is. An agent that can explain the client back to you in five lines is ready to work on the site. One that can't would have guessed.
What should never go in the client folder?
Anything that would hurt the client if the folder were copied, shared or read by the wrong tool:
- Passwords and login codes. Use a password manager. Where a service allows it, invite yourself to the client's account with your own login instead of sharing theirs.
- Card numbers and bank details. They belong in the payment account, not in a note.
- Private details about the owner's customers, such as a salon's client list, unless the work needs them and the client has agreed.
The folder can and should say which accounts exist and who owns each one: "Domain: in the client's name. Google Business Profile: client is owner, we're a manager." Keeping client accounts in the owner's name is the rule that makes that list short and safe.
How do you keep the client's history useful?
Date everything, and never rewrite the past. One file per call, decision or approval in history/, with the date at the start of the file name:
history/
2026-03-02-kickoff-call.md
2026-03-10-homepage-draft-approved.md
2026-03-18-client-asked-for-gift-card-page.md
When the client says "I thought we agreed...", you open one file instead of scrolling a month of email.
Approvals deserve their own note: what was approved, which version, by whom, and when. That's also the line between what your agent may do on its own and what waits for you. AI agent approval sets out which things should never go to a client without your say-so, and approving what your agent sends shows how that looks in practice.
Review the folder once a week, as part of keeping track of clients and promises: read the latest history notes, update the brief if something changed, and move anything the client sent into content/.
Where Volant fits
Volant keeps each client's notes the same way. Your notes on each client and their brand are ordinary text files you can open in any text editor, and your agent reads them when it works on that client. Volant's own records about the client sit beside those notes, so the brief and the history stay in one place. If you ever move on, the notes are still just files in a folder.