Here’s how it usually goes.

Your company rolls out Claude. There’s maybe a training, maybe just a Slack message with a login and a cheerful “go explore.” You open it between two meetings, type “write a follow-up to the client,” and get back something fine. Generic, but fine. You tweak it, you send it, you close the tab. Three weeks later you’re still using a tool that could rewrite your entire job like a slightly fancier search bar.

That’s not on you. Claude showed up knowing nothing about your company, your role, or how you sound, so it wrote for the average person doing the average version of your job. You never told it who it was working for.

Here’s the trade: a little focused time up front, and every task after that comes back in your voice, aimed at your world. People give it fancy names. I’ll call it what it is: your second brain. Here’s exactly how to build one, then how to build Skills that run whole jobs on top of it.

Part 1: Build the brain

What a second brain actually is

It’s a small, reusable set of context about you and your work that Claude reads before it does anything, so it stops guessing and starts sounding like it actually works with you. Build it once. Everything you make after that starts from it.

Same idea my other company, Numeriq Advisors, installs for whole businesses: one “AI Brain” that holds how the best people work and feeds every function. This is the personal version. One person, one afternoon.

Why this is the whole ballgame

Once the brain exists, everything you make gets sharper, because Claude finally has the context it was missing.

The same ten-word request (“draft the board update”) now comes back in your voice, aimed at your board, using your numbers, and you didn’t write a single instruction to get there. The people getting scary-good work out of Claude aren’t writing better prompts. They just did this part first.

Engine vs. storage (the thing that trips everyone up)

One idea saves a lot of confusion. People mix up how Claude works with where your brain lives. Two different things.

The engine is how Claude works. Chat is quick back-and-forth. Cowork is the heavy engine that runs long, multi-step work on your files. It’s optional, and your admin decides whether you have it.

The storage is where your brain lives, and it’s the part that matters:

• A Project holds your real second brain (your company, role, voice, and collateral). It’s yours by default, and it’s the one thing you can share with your team.

• Personal Preferences is a tiny always-on note about how you like to work, riding along in every chat.

The freeing part: your brain lives in a Project plus that short note, not in some special mode. So it works the same on Pro, Team, or Enterprise, with or without Cowork. Cowork is just a stronger engine you can point at the same brain later.

If you’re using Cowork

If your company gave you the Cowork desktop app and that’s where you’re working, everything here still holds — same three files, same brain. The only thing that changes is where the files live.

Cowork works from a folder on your computer, not from a Project’s knowledge base in the browser. So in Cowork, your brain is a real folder (call it “My Work Brain”) with your three files saved inside it as plain documents, plus your collateral (brand, templates, a couple of finished pieces) dropped in as files. Point Cowork at that folder and it reads them the same way the browser reads a Project.

Two things to know so you don’t get tripped up:

• Build the three files once, keep the master copies somewhere you control. Paste them into a browser Project when you’re on the web; keep them in your Cowork folder when you’re on the desktop. They don’t sync on their own — you’re keeping two homes for the same files.

• Your always-on note lives in Settings and rides along in both places, so you only write it once (Step 2).

So when you hit “Make your Project” below, a Cowork-only reader can make a folder instead. The interviews, the files, and the test are identical either way.

Your brain is three files

Don’t overbuild this. The brain is three short pieces of writing, and they all live in your Project:

1. My Company. What your company does, how it wins, who it serves, the brand and tone, the goals this year. True no matter who holds your job.

2. My Role. What you own, your day-to-day, who you answer to, what you’re measured on. Your specific seat.

3. My Voice. How you write and sound, plus the AI clichés you never want going out with your name on them.

You’ll build all three by talking, not typing. Here’s the whole thing, in order.

Step 1: Open Claude and Wispr Flow (5 min)

Claude. The browser works fine for all of this. If your company also gave you the desktop app with a Cowork tab, great, but you don’t need it to build your brain.

Wispr Flow. This is the part most people skip, and it’s the part that makes the whole thing work when you’re busy. Wispr Flow is a dictation tool: hold one key, talk, and your words show up wherever your cursor is. The free tier is plenty.

Why it matters: the setup below works by Claude interviewing you, and good answers are the whole point. You talk at about 150 words a minute. You type at maybe 40 (and if you’re me, half of those are typos). You also think differently out loud. You ramble, you give the real example, you say the thing you’d never bother typing. That mess is what makes your brain sound like you. Install it first, then talk your answers.

Can’t install it, because your work laptop is locked down? No problem. Type your answers instead, and just make them long and specific. Talking is the easy mode here, not a requirement. Don’t let a missing app stop you from building the brain.

Step 2: Set your always-on note (2 min)

Open Settings, go to your Profile, and find the box that asks what Claude should keep in mind about you. This note rides into every chat, so keep it short and lane-neutral. It’s your working style, not your company. Paste something like:

Be concise and direct. Skip the preamble and filler.
If a request is unclear, ask me one or two sharp questions before you start. Don't guess.
Use plain language, no corporate buzzwords.
When you write in my voice, follow the My Voice file in my project.

That last line is the trick: your actual voice lives in the Project (Step 4), and this note just points to it. Keep company-specific detail out of here, because this note also shows up in your personal chats.

Step 3: Make your Project (2 min)

In the left sidebar, click Projects, then New project. Name it something like “My Work Brain.” You’ll see two boxes: instructions (always loaded) and a knowledge base (files it pulls from when relevant). That’s where the three files go next.

No local folder to build or maintain. And because a Project is a normal feature on every paid plan, you can set this up yourself even if IT has locked down everything else.

Step 4: Build the three files (10-15 min each)

Open a normal chat and paste one prompt at a time. Each runs a short interview: Claude asks, you talk, it writes the file. Then you drop the result into your Project. Do them in order. Break between them if you need to.

Prompt 1: My Company

Help me write the "My Company" page of my Claude second brain. It lives in my
Project and loads before every task, so keep it short and useful — under a page.

Don't write anything yet. Interview me first. Ask me three or four plain
questions at a time, then wait for my answers before the next batch. I'll answer
out loud, so I might ramble — pull out what matters and leave the rest.

Ask me things like:
- If you had to explain what we do to a smart twelve-year-old, what would you say?
- What do people actually pay us for?
- Who is our best-fit customer, and what keeps them up at night?
- When someone picks us over the other option, why do they say yes?
- If our company were a person, how would it talk?
- What is the one big thing we're trying to pull off this year?

If an answer comes out fuzzy, ask me for a real example instead of moving on.
Never fill a gap by guessing — ask me.

When you have enough, write the page: plain sentences, a few short lists, no
filler and no buzzwords. Then give me two versions — a four-to-five-line summary
to paste into my Project instructions, and the full page to save as a knowledge
file.

Tip: paste your company website link into the chat so Claude can pull from it while it interviews you.

Prompt 2: My Role

Now help me write the "My Role" page of the same second brain. The company page
already covers the company — don't repeat any of it. This one is only about my job.

Same as before: interview me first. Ask three or four simple questions at a time
and wait for my answers. I'll talk it through, so listen for what matters.

Ask me things like:
- What's your job title, and what does that actually mean day to day?
- What lands on your desk that really only you can handle?
- Who do you work with most, and who do you report to?
- How does your boss decide whether you had a good month?
- What do you end up making again and again — the same reports, emails, or docs?
- When you hand something off, how do you know it's actually good?

If an answer is vague, ask me for the specific real example. Ask, don't assume.

Then write the page: tight, mostly short lists, under a page. Give me the
four-to-five-line summary for my Project instructions and the full page to save.

Prompt 3: My Voice (spend the most time here)

This is the file that stops Claude sounding like a stranger who skimmed your LinkedIn once. It has Claude study your actual writing, not just ask about it.

Now the most important page: "My Voice." This is what makes Claude write like me
instead of like a robot. Take your time here.

First, learn how I write. Ask me to paste in three things I wrote and liked — an
email, a text, a note, a post, anything real. Read them and tell me what you
notice, with proof from the samples: how long my sentences run, how formal or
loose I am, the words I reach for, how I start, how I sign off, how I handle bad
news. Don't flatter me. Just tell me what's actually there.

Then ask me a few simple questions, a few at a time:
- Who am I usually writing to, and how do I want them to feel?
- Am I more buttoned-up or more casual? Does it change with who I'm writing to?
- Any words or phrases that feel like me?
- Any words or phrases that make me cringe?
- If a coworker read something with no name on it, what would tell them it's mine?

Then write the page in three parts:

1. How I sound — a real description in my own style, pulled from my samples and
answers. Include the three-to-five moves I make every time, and a plain do / don't
list. Be specific, not generic.

2. Words to never use — the AI giveaways to keep out of my writing. Start with
these and let me add my own:
- the "it's not X, it's Y" flip, in all its forms
- puffed-up words: leverage, delve, unlock, tapestry, robust, seamless, elevate,
  game-changer, and their cousins
- the tidy group-of-three ("faster, smarter, better")
- tired openers: "In today's world...", "It's worth noting that..."
- long dashes stacked in one sentence, and every line the same length

3. Before and after — two quick rewrites that take a stiff, generic line and turn
it into how I'd actually say it, so future-me and Claude can see the difference.

Keep it to about a page and a half. Give me the short version for my Project
instructions and the full version to save as a knowledge file.

Step 5: Feed it your real work (5 min)

The three files describe your world. Your collateral shows it. Upload the real material into your Project’s knowledge base, and your output stops being about your company and starts looking like it came from it: your brand (logos, exact colors, style guide), your templates (a board deck, a proposal, the monthly report), and two or three finished pieces you were proud of, so Claude knows the bar.

Then point Claude at them: “Draft the Q3 board deck. Match the template and colors in this project’s knowledge.”

Step 6: Test it (2 min)

Open a fresh chat inside your Project and ask for something real, with no instructions: “Draft my weekly update to my manager.” If the brain’s working, it comes back in your voice, aimed at the right person, using your priorities. If something’s off, open the one file that got it wrong and fix that file. That’s the whole maintenance loop.

When the test comes back wrong

The first try usually isn’t perfect, and that’s fine. Almost every miss traces back to one file. Match the symptom to the fix:

• Still sounds generic or stiff → your My Voice file is too thin. Add real writing samples and specific do’s and don’ts, not adjectives.

• It made up a number, name, or quote → it’s missing your real material. Upload the actual document into your knowledge and tell it to use only what’s there.

• Too long, or too formal → say so in My Voice, add a “keep it short” line, and give it one before-and-after example.

• Describes your company but doesn’t feel like it came from you → that’s a My Role gap. Spell out what you own and who you’re writing for.

• Ignored your template or brand colors → point it straight at the file: “match the template and colors in this project’s knowledge.”

Fix the one file, run the same test again, and move on. You’re not chasing perfect — you’re chasing “sounds like me, mostly.” That’s the file doing its job.

Skip these four mistakes

1. Dumping confidential data. Keep it to what you’d already put in a normal email. No client data, no secrets, unless your company’s policy clearly allows it.

2. Ten-page files. If a file gets bloated, Claude skims it instead of reading it. Lean or it stops working.

3. Company voice in the always-on note. Personal Preferences applies everywhere, including your personal chats. Keep the company and role stuff in the Project.

4. Giving short answers. Short answers make a generic brain. However you answer, go long and give the real example — that’s what makes it sound like you, not whether you typed or talked.

Don’t overload it

More context isn’t always better. A tight brain makes Claude sharp. A bloated one makes it repetitive, because it leans on the same loaded material for every answer. Fine when you want the same board deck every quarter. A problem when you want it to think with you on something new.

Keep the always-on layers lean, and pull in heavy collateral only when the task calls for it.

The corporate stuff (30 seconds)

Greyed-out buttons? If Cowork, a connector, or a setting is missing, that’s your admin, not you. The Project path here needs none of it.

Your data? On Team or Enterprise, your company’s setup generally keeps your work out of model training, so you can add your work context with reasonable confidence (still follow your own policy). On a personal Pro account, you can turn training off yourself under Settings → Privacy.

Keep it fresh

Your job changes, your goals move, new AI clichés start driving you up the wall. Once a quarter, spend ten minutes rereading your three files. Cut what’s stale, update this year’s goals, add new phrases to the banned list, swap in a fresher writing sample. A brain you never update ages like milk.

Reuse it across Projects

The moment you spin up a second Project (a client, a launch, a new team), you’ll want the same you inside it. Here’s the rule that trips people up: a Project’s files stay in that Project. They don’t sync.

So keep your three About Me files saved somewhere central, and re-drop them into each new Project’s knowledge base, with the short summary pasted into its instructions. Then add only what’s specific to that project on top. Your Personal Preferences already ride along everywhere, so the light layer is automatic. Build the About Me once, and stamp it into every new workspace.

Part 2: Turn your brain into Skills

You built the brain. Now stop re-explaining the same tasks.

A Skill is a saved capability you trigger with a slash command, for the work you do again and again. Your brain tells Claude who you are. A Skill tells it how to do a specific job the way you like it, every time, without re-explaining. And because the Skill runs on top of your second brain, it already knows your voice, your brand, and your numbers.

Where Skills live

Two places. You fire a Skill by typing / in any chat and picking it from the menu. You manage them (create, edit, turn on and off) under Customize → Skills. If you don’t see Customize, your company may have it switched off. You can still keep a Skill’s instructions in a note and paste them when you need them.

How to make one

You don’t write code. You describe what you want, out loud, and Claude builds it.

Two ways in:

1. Run /skill-creator and talk it through (Wispr Flow again). Tell it what the skill should do, what to ask you for, and what “good” looks like.

2. Save a win. Any time Claude nails a task, say “save this as a skill.” It turns that chat into a reusable command.

The whole trick is the description. You’re not programming. You’re briefing a sharp assistant on a job you want done the same way every time.

Four skills worth building first

Here’s what each one does, and the plain-language brief you’d hand /skill-creator to build it. Swap in your own details.

1. Friday Recap Email · /friday-recap Turns your week into a polished recap email to your boss.

Build a skill called friday-recap. When I run it, ask me what I worked on this week: wins, what moved forward, anything that slipped, and what’s next. Then write a tight recap email to my manager in my voice (follow my Voice file): a one-line summary up top, then wins, in-progress, and next week, with any blockers flagged. Keep it skimmable. Only use what I actually gave you, don’t inflate.

2. Friday End-of-Day Recap · /eod-recap A quick, informal end-of-week wrap-up for you or your team.

Build a skill called eod-recap. When I run it, ask how the week went and what I got done. Then give me a short, casual end-of-day recap I can drop in Slack or keep for myself: what shipped, what’s still open, and where I’m picking up Monday. Loose and human, not a formal report. Same facts as my boss recap, lighter tone.

3. Content Creation Engine · /content-engine Drafts posts in your voice, and a first-draft press release when you need one.

Build a skill called content-engine. When I run it, ask what I want to make (a social post, or a PR / press release draft), what happened, and who it’s for. For a post: draft it in my voice (follow my Voice file), lead with a real hook, keep it tight, give me two versions. For a PR draft: use my company’s tone from my project, with a headline, a strong first paragraph, a quote slot, and a boilerplate at the end. Only the facts I gave you, don’t invent numbers or quotes.

4. Recurring Meeting Agenda · /meeting-agenda Builds the agenda for your recurring meeting, in your format, every time.

Build a skill called meeting-agenda. When I run it, ask which meeting it’s for, who’s attending, and what’s on my plate this week. Then build the agenda in my usual format: carry over open items from last time, group by topic, and leave space for notes and action items. Keep it tight enough to actually get through. Match how I run that meeting.

The rest of the menu (build these the same way when you need them): /meeting-prep, /board-update, /exec-summary, /competitor-onepager, /email-reply. Same move every time: describe what you want, let Claude build it, trigger it with /.

One rule: start with one. Pick the task you do most, build that skill, use it for a week. Then add the next.

Your next moves

Take one sitting this week and build the brain: Steps 1 through 4 (through your first file) together, My Role and My Voice on two more short breaks, then feed it your collateral and test it. A working second brain by Friday. Then build one Skill, the task you do most, and watch a week of that job get shorter.

That’s the system: a brain that knows you, and Skills that run your repeat work on top of it. Next time I’ll show you how to stack Skills into full workflows, and how a team can share one brain instead of everyone building their own.

Save this. Send it to the colleague whose company just gave them Claude and who still hasn’t opened it.

I’ll see you after carpool.

AI After Carpool · Build your second brain