Before You Start
You do not have a prompt problem
Almost everyone who feels let down by AI is running it completely raw. No instructions, no tools connected, no memory of anything they said last week. Then they go hunting for a better prompt to fix it.
A great prompt on a raw Claude loses to an average prompt on a set-up one. Every time. Because the set-up one already knows who you are, it can reach into the apps where your work actually lives, and it remembers the last six months of your context.
That is all three of these things. Knows you, can reach your tools, remembers. Nothing here is a trick and nothing here expires next month when a new model drops.
The Three
About 30 minutes, in this order
The order matters. Each one gives Claude something it needs for the next. If you only get through the first one today, you are still ahead of nearly everybody.
Global instructions
10 minClaude interviews you, then writes the block that makes every future chat start knowing who you are. You stop re-explaining yourself.
Connectors
5 minGmail, Calendar, Notion, Slack, whatever you actually live in. This is the difference between Claude talking about your work and Claude doing it.
The second brain folder
15 minFive files on your computer that hold everything about your work and your life. Claude writes them for you, then edits them forever.
Thing 1 — Prompt 1
Stop re-explaining yourself
Global instructions are a block of text Claude reads at the top of every single chat, before you type anything. Most people either leave it blank or write two sad lines in it.
You are not going to write it. Claude is. Paste this into a fresh chat and it interviews you one question at a time, then hands you a finished block built out of your real answers.
Answer honestly and do not rush it. Fifteen minutes of good answers here pays out in every conversation you have for the next year.
Copy-paste prompt
You are going to help me set up Claude so that every answer you give me from now on is tailored to me, my work, and how I think. We're going to do this as an interview, not a form. Do not skip ahead, and do not dump a giant list of questions at once. The goal: by the end of this chat, you write me a personal-instructions block I can paste into Claude's settings, so every future conversation already knows who I am, what I do, what I'm trying to get done, and how I like you to respond. Rules for the interview: - Ask me ONE question at a time, then wait for my answer before the next one. - Start broad, then go deeper based on what I say. Follow the thread. - If an answer is vague, ask a follow-up until it's concrete. Push me for specifics: real numbers, real names, real examples. - Keep it conversational, like a smart friend getting to know me. No survey energy. - Roughly 12 to 18 questions total. Don't stop early just because you think you have enough. Cover these areas, but in a natural order, not as a checklist you read out loud: 1. Who I am. Name, where I live, what I do for work, my role, how long I've done it. 2. My work. The business or job in detail. What I actually produce. Who my customers or team are. How I make money or get judged. 3. My goals. What I'm trying to hit in the next 90 days and the next year. What "winning" looks like for me. 4. My week. How I actually spend my time. What eats my hours. What I dread. What I'm best at. 5. My skills and gaps. What I'm great at, what I'm weak at, what I'd love to just hand off to you. 6. How I communicate. My tone, the words I use, the words I hate, examples of writing that sounds like me. 7. How I want you to respond. Long or short. Direct or gentle. Push back on me or agree. Ask before acting or just go. 8. Anything else a great assistant would kill to know about me. When you have enough, say: "I have what I need. Here's your personal-instructions block." Then output a clean, copy-paste block, written in the first person as instructions TO you (Claude), that captures: - Who I am and what I do (tight and factual) - What I'm working toward - How I want you to communicate with me - Any standing rules (things to always do, things to never do) - The one instruction I explain below Add this line to the block, word for word: "Address me by my first name in every single reply." That's a reliability trick. The moment you stop using my name, it's my signal the chat has gotten too long and I should start a fresh one. Keep the final block under 400 words. Plain language, no fluff. This is the thing I'm going to live with in every single chat, so make it count. Start now with your first question. Just one.
Where the block goes
Claude desktop app, Settings, then Customize, then the global instructions box. Some versions call it Personalization or Profile. Paste, save, done. It applies to every new chat from that second on.
Checkpoint
Open a brand new chat and just say "hey." It should greet you by your first name and reference something real about your work. If it does not, the block never saved.
The name trick
The block tells Claude to use your first name in every reply. That is not vanity, it is a warning light. The moment Claude stops using your name, the chat has run long enough that quality is slipping, and it is your cue to start a fresh one.
Thing 2
Give it hands
Out of the box Claude can only talk to you. Connectors are what let it go do things in the apps you already use, without you leaving the desktop app to be the middleman between every step.
Settings, then Customize, then Connectors. Turn on the ones you actually open every day. If a tool is not in the built-in list, it is almost certainly available as an MCP, which is just the standard way any tool plugs into any AI. Add it there.
Do not connect fourteen things because you can. Connect the three or four you live in, and connect them today.
Gmail
Inbox triage, drafted replies, follow-ups it writes and leaves waiting for your approval.
Calendar
Real answers about your week. Meeting prep that pulls the actual attendees and the actual agenda.
Notion or Drive
Reads context out of the docs you already keep, and writes new ones back where your team will find them.
Slack
Catches you up on what you missed and drafts the reply in the thread instead of in a chat window.
The tool you use most
HubSpot, Linear, Figma, Higgsfield, Stripe. The one you open first every morning is the one worth connecting first.
Not on the list?
Search the tool name and "MCP server." One exists for most tools now. If it truly does not, tell Claude the tool and ask for the next-best path.
Checkpoint
Ask: "What is on my calendar tomorrow?" It should answer with your actual meetings. If it offers to help you check, the connector is not live yet.
Thing 3 — Prompt 2
Build the second brain
Global instructions are a paragraph about you. A second brain is the whole file cabinet, and it is the piece almost nobody has.
Make a new folder on your computer. Call it second-brain. Leave it empty. Then open the Claude desktop app, switch to Cowork, click add folder, and select the one you just made.
That is the part people miss. In Cowork, Claude can read and write real files on your computer, so it does not just describe a second brain to you, it builds one and then keeps editing it every time something about your life changes.
Five files go in there. Claude writes all five for you off the back of one interview.
START-HERE.md
The map, and the standing orders. Tells every future version of Claude to read the other four files first and to update them the moment a fact changes.
about-me.md
Who you are, what you are good and bad at, how you talk, and how you want Claude to talk to you.
work.md
The job or business in real detail. What you produce, who pays you, every active project with its status, and your 90-day numbers.
people.md
One block per person who matters, work and personal. The 2 or 3 things Claude needs to help you with anything involving them.
routines.md
Everything that repeats in your week, plus a ranked list of which ones Claude could take off your plate first.
Paste this into that Cowork session with the folder attached. It interviews you about both your work and your life, then writes all five files with your real details in them.
Copy-paste prompt
You have a folder attached to this conversation. It is my AI second brain, and right now it is empty. By the end of this chat there will be 5 real files in it, filled with real details about me, so that every future conversation we have starts with you already knowing my whole situation. Before anything else: confirm you can see the folder and tell me its name. If you cannot see it, stop and tell me, because nothing below will work. We do this in two parts. Part 1 is an interview. Part 2 is you writing the files. Do not skip to Part 2. PART 1 — THE INTERVIEW Rules: - Ask me ONE question at a time, then wait for my answer. - Follow the thread. If an answer is interesting or vague, dig into it before moving on. - Push me for specifics. Real names, real numbers, real deadlines, real dollar amounts. - Talk like a sharp friend who is genuinely curious, not a form. - Roughly 15 to 20 questions. Do not stop early because you think you have enough. Cover both sides of my life, in a natural order, not as a checklist you read out loud: Work - What I actually do, and the current state of it. Projects, clients, revenue, deadlines, what is on fire right now. - The people in my work world. Team, clients, partners, my boss, and the one thing that matters about each. - Where my important stuff already lives. Which apps and tools I am in every day. - The work that repeats. What I do every week or every month without fail. - My numbers. What I am chasing in the next 90 days and what winning looks like. Personal - The areas of my life outside work that matter. Health, family, money, home, learning, hobbies. - The people in my personal life and what matters about each. - What I am trying to build or break. Habits, goals, things I keep starting over. - What is stressing me out or quietly getting dropped. When you have enough, say: "I have what I need. Building your second brain now." Then go straight into Part 2 without waiting for me. PART 2 — BUILD THE 5 FILES Create exactly these 5 files in the attached folder. Use these exact filenames. 1. START-HERE.md The map and the standing orders. It explains what each of the other 4 files holds and tells your future self how to use this brain. Write that part as direct instructions to you, not notes to me. Include these lines close to word for word: "Read every file in this folder before answering anything that involves my work, my week, or the people in my life." "When I tell you something that changes a fact in these files, update the file in the same reply. Do not wait to be asked." "When a file gets long enough to be hard to scan, tell me and propose how to split it." Also put the date it was built and a one-paragraph summary of who I am, so a fresh chat gets the gist from this file alone. 2. about-me.md Who I am. Name, where I live, what I do, my story in a short paragraph, what I am good at, what I am bad at, what I want people to hand off to me and what I want to hand off. How I communicate: my tone, the words I use, the words I hate. How I want you to talk to me: length, directness, whether to push back, whether to ask before acting. 3. work.md My job or business in real detail. What I produce. Who pays me or judges me. How the money actually works. Every active project as its own short block with status, what is next, and what is blocked. My 90-day targets with the real numbers attached. 4. people.md One short block per person that matters, work and personal mixed, each with their name, their relationship to me, and the 2 or 3 things you would need to know to help me with anything involving them. Preferences, history, sore spots, what they are working on. 5. routines.md Every recurring thing in my life, work and personal, with how often it happens and roughly how long it takes me. At the bottom, a section called "Automation candidates" where you rank the top 5 of these by how much time they eat and how cleanly you could do them for me. Rules for writing the files: - Fill them with the actual details I gave you in this conversation. No placeholders, no [brackets], no "add your info here." - If something is missing and it matters, ask me one more question instead of guessing. - Plain markdown, short lines, headings and bullets. These are files you will re-read a hundred times, so make them fast to scan, not pretty. - Do not invent anything I did not tell you. If you are unsure whether I said something, ask. When all 5 files exist, show me the folder contents, then tell me the single habit that keeps a second brain alive and what will happen if I skip it.
Checkpoint
Start a fresh Cowork session, attach the folder, and ask: "Based on my routines file, what should I hand off to you first?" It should name a real task from your real week.
Attach it every time
The folder is not automatic. Every new Cowork session, attach it again. It takes two clicks and it is the difference between Claude guessing about your week and Claude knowing it.
That Is All Three
What to do from here
Claude now knows who you are, can reach into the tools where your work lives, and has a memory of your whole situation sitting on your hard drive. That is a system instead of a chatbot, and you never have to build it again.
Three things to carry forward:
Tell Claude when something changes. A second brain dies the week you stop updating it, and it only takes one sentence in whatever chat you are already in.
When Claude gets something wrong about you, that is a missing file, not a bad model. Add the fact and move on.
The second time you explain the same task, stop and turn it into a skill instead. That is the whole next level, and you are now set up for it.
Thirty minutes once, and every chat after it starts from a completely different place.