AI Guides › Playbooks
By Nigel Guy · 6 min read
Most people brief their AI from scratch every morning. They paste the same paragraph about their business, their clients and their tone of voice, get a decent answer, close the tab, and do it all again tomorrow. It feels fine because each individual answer is fine. The cost is the daily re-explaining, and the quiet drift between versions of "who I am" scattered across chats.
The rule: write your context down once, in one place you own, and point every AI tool at that place instead of at your memory.
Strip away the productivity-influencer packaging and it is two things: a small set of files that describe your work (who you serve, how you decide, what good looks like, what is currently live), and a way of getting those files in front of an AI at the start of a task. The tool matters far less than the discipline of keeping the files true.
Three rungs, easiest first. Each rung uses the same files; only the delivery changes.
| Rung | What it is | Cost at time of writing | Best for | Catch |
|---|---|---|---|---|
| 1. Claude Project | A workspace on claude.ai with a knowledge base and project instructions | Free plan allows up to five projects; Pro is listed at $17 a month billed annually or $20 monthly (about £13 to £15 at time of writing; check the £ price at checkout) | Getting started today, one topic at a time | Lives inside Claude only; you maintain it by uploading |
| 2. Obsidian vault | A folder of Markdown notes with links and search | Core app free, including commercial use; optional Sync $4 to $5 a month (about £3 to £4 at time of writing) | Writing and linking notes you read yourself | A note app, not an AI delivery system; you still have to feed the AI |
| 3. Plain folder plus Claude Code | The same Markdown files, read directly by an AI that works in the folder | Claude Code is included in Claude Pro per Anthropic's pricing page; check current plans | Context that several tasks and tools can share | Needs comfort with a terminal or desktop app, and a habit of tidying |
Anthropic's help centre describes Projects as self-contained workspaces: each has its own chat history, a knowledge base of uploaded files, and project instructions that shape Claude's behaviour inside that space. Paid plans add retrieval (RAG), which Anthropic says expands capacity by up to 10x as you approach the context limit.
Two traps. Information is not shared between chats unless it is in the knowledge base, so a good decision made in a chat vanishes unless you copy it into a file. And the project name and description are not read by Claude, so do not put context there.
Obsidian stores notes as Markdown plain text in a "vault", which its documentation defines as a folder on your local file system, including subfolders. That is the important sentence: an Obsidian vault is already a plain folder. You get links between notes, a graph view and good search, and nothing is locked in. Use it if you like reading and connecting notes yourself. Do not assume that installing it gives your AI anything; Obsidian does not brief Claude for you.
This is the most powerful rung because the AI works where the files live. Claude Code reads a CLAUDE.md file at the start of each session. Per Anthropic's documentation:
~/.claude/CLAUDE.md holds personal preferences across all projects; a CLAUDE.md in the project folder holds that project's context.@path/to/file inside a CLAUDE.md imports another file, up to four hops deep. Imports organise a long file but do not reduce its context cost, as imported files load at launch./init drafts a starting file for a codebase. It is aimed at code projects, so for a business folder you will write your own.A workable layout:
brain/
CLAUDE.md (the index: who I am, how to work, where things are)
about-me.md
clients/
voice-and-style.md
decisions-log.md
current-priorities.md
Keep CLAUDE.md short and have it point to the rest. Anthropic notes the file loads in full every session, so an index beats an encyclopedia.
Fill in the bracketed parts, run it in a Claude chat, then save the result as about-me.md.
You are an editor helping me write a one-page context file that an AI assistant will read before every task.
My work: [WHAT_YOU_DO_AND_FOR_WHOM]
My tone of voice: [THREE_WORDS_AND_ONE_EXAMPLE_SENTENCE]
Things the assistant must never do: [LIST]
Current priorities this month: [LIST]
Do this in order:
1. Ask me up to five questions about anything above that is vague or missing. Wait for my answers. Do not guess.
2. Draft the file under these headings: Who I serve, How I want to sound, Hard rules, Current priorities, Words I avoid.
3. Keep it under 400 words, in plain sentences, with no flattery and no invented facts.
4. Before showing it, check each line is something I told you, and mark anything you inferred with [CHECK].
Start on rung 1 with one project and five files. Move to rung 3 only when you notice you are uploading the same files into several projects, or when you want one source of truth across tools. Obsidian is optional at every stage; add it if you enjoy writing notes, not as a prerequisite.
A freelance bookkeeper in Leeds starts with a Project called "Client emails" holding her tone sheet, a fee schedule and three anonymised sample replies. After a month she notices she has uploaded the tone sheet to three projects and edited two copies. She moves the files into one folder, writes a 40-line CLAUDE.md that imports the tone sheet and fee schedule, and edits each file once from then on.
CLAUDE.md. If it passes 200 lines, split it.Not verified: any file-size limits for Project uploads, which Anthropic's pages did not state.