AI Guides › Step-by-step guides
By Nigel Guy · 8 min read
Most people build a second brain the wrong way round. They pick the tool first (usually Obsidian), spend a weekend on plugins and tags, then find the AI they actually work with can't see any of it. It feels productive because it looks organised, while you keep pasting the same background into every new chat.
The rule: a second brain is one place where your working information lives in a form an AI can read without you re-explaining it. Pick the simplest container that does that, and spend your effort on what goes in it, not on the container.
Three containers, least effort first: a Claude Project, a plain folder of text files, and Obsidian, which is the folder with a nicer viewer.
| What | Do you need it? | Cost at time of writing |
|---|---|---|
| A Claude account | Yes, for option A | Free plan includes Projects, capped at five. Pro is listed at US$20 a month (US$17 a month billed annually); the UK price is shown in pounds at checkout and includes VAT, so check it there |
| A computer with a normal file system | Yes, for option B | Nothing extra |
| Claude Desktop with Cowork, or Claude Code | Only if you want Claude to read the folder directly | Cowork needs a paid plan (Pro, Max, Team or Enterprise); Claude Code is included with Pro |
| Obsidian | Optional | The app is free, including for work use. Sync is listed at US$4 a month billed annually; you don't need it |
Open a blank note and list the five to ten things you re-explain to AI most often. Typical examples: who your clients are, how you price, your writing voice, current projects and their status, recurring templates, decisions you've already made and why.
Note where each lives today. Anything that lives only in your head gets written down first; that's the real work. Cross out anything that changes daily, such as today's to-do list. A second brain holds stable context, not a live feed.
Use this if you mostly work in Claude on the web or the app and your context fits in a handful of documents.
Know the limits. Project files can be up to 30MB each, and only their text is extracted (PDFs excepted). On paid plans Claude switches to a retrieval mode when project knowledge gets large; the free plan doesn't. Chats in a project don't see each other, so add anything worth keeping to the project knowledge yourself.
Use this prompt in a new chat inside the project to write the instructions and decide what to upload. Fill in the area of work and paste your Step 1 inventory.
You are helping me set up a Claude Project so I never have to re-explain my background in new chats.
Area this project covers: [AREA_OF_WORK_OR_LIFE]
My inventory of things I keep re-explaining: [PASTE_STEP_1_INVENTORY]
Documents I'm thinking of uploading: [LIST_OF_CANDIDATE_FILES]
Work in this order:
1. Ask me up to five questions, one at a time, about whatever you would still need to know to be useful in this area without further explanation. Stop and wait for each answer. If my inventory already answers something, don't ask it.
2. Draft project instructions of no more than 300 words: who I am in this context, what I usually ask for, the standards a good answer meets, and anything you must never assume.
3. Sort my candidate documents into a table with three columns: Upload (genuinely reused context), Summarise first (useful but long or messy), Leave out (out of date, one-off, or duplicated). Give a one-line reason for each.
Rules: do not invent facts about me or my work. If an answer is missing, ask rather than fill the gap. Use British English.
Before you reply with the final draft, check: is every instruction something I actually told you, and is the document list short enough that it won't bury the important files?
Use this if you switch between AI tools, want your context to outlive any one product, or want Claude to read and update the files itself.
second-brain, somewhere that's backed up.about/ (you, your voice, your business), work/ (one subfolder per active client or project), reference/ (prices, templates, processes), archive/..md). One topic per file, a clear filename, a date at the top when it matters.If you use Claude Code, name that map file CLAUDE.md: Claude Code loads it at the start of every session. Anthropic's docs recommend keeping it under 200 lines, and /init will generate a first draft for you. Claude Code also reads AGENTS.md, a name other AI tools recognise. In Claude Desktop, Cowork can read and write files in a local folder you connect.
This prompt designs the folder and drafts the map file. Fill in what you work on.
You are designing a plain-folder knowledge base that any AI assistant can read cold.
What I work on: [YOUR_ROLES_CLIENTS_AND_PROJECTS]
Tools I use with AI: [E.G. CLAUDE CODE, COWORK, CHATGPT]
Things I keep re-explaining: [PASTE_STEP_1_INVENTORY]
Steps:
1. If anything above is vague or missing, ask me about it first. Don't guess my work.
2. Propose a folder structure no deeper than two levels, shown as a tree. Justify each top-level folder in one line.
3. List the first five files I should write, in priority order, with three to five bullet points on what goes in each.
4. Draft the top-level map file (under 120 lines, Markdown): what this folder is, what each folder holds, which file to read first, naming conventions, and what an AI must not change without asking.
Constraints: no tags, no plugins, no databases. Plain text only. British English.
Self-check before answering: could someone keep this up in six months with ten minutes a week? If not, cut it down.
An Obsidian vault is, in Obsidian's own words, a folder on your file system, and the notes are Markdown plain-text files. So if you built Step 3, you already have something Obsidian can open: choose to open an existing folder as a vault and point it at second-brain. Obsidian adds links between notes, search and a pleasant editor. It also adds a hidden .obsidian settings folder, which you can ignore.
Add it if you write lots of notes by hand. Don't adopt it for the graph view or plugins: the AI reads the files, not the visuals.
Archive what's finished, update what changed, and add one thing you re-explained that week.
Start a fresh chat (or a new session in the folder) and ask three questions you'd normally have to brief: "Who is my biggest client and what do they care about?", "What's my standard day rate?", "Draft a reply in my usual tone." Then: