🔵 Building. Last of the context issues, builds on 1 through 7. If you’ve ever pasted an email thread into Claude just to get help replying, today ends that.

New here? Every issue builds on the one before it. Catch up: 1 · 2 · 3 · 4 · 5 · 6

One prompt. Claude read the calendar, searched the inbox, wrote the brief. Nobody pasted anything.

Last week I ended with a promise: say “find the meeting notes” and Claude goes and looks. Today we make that real.

You know the old routine. Open Gmail. Copy the thread. Paste it into Claude. Copy the draft. Paste it back. You were the connector.

Connectors let Claude reach into your apps directly:

• Gmail: search your real inbox, draft replies.

• Google Calendar: read your week, move things on it.

• Google Drive: fetch “that doc from March” by name.

Today: connect all three, test each, run one prompt that uses them together. 20 minutes. Works on every plan, Free included. Second tab open, click along.

Last Week’s Homework

The File Triathlon: a PDF, a spreadsheet, a whiteboard photo, each fed to Claude by hand. I asked for your oldest “deal with it later” file. Shout-out to REPLACE-WITH-READER-NAME, whose REPLACE-WITH-READER-STORY.

Same idea today, different courier. Last week you brought the file. Now Claude goes and gets it.

Part 1: Three Facts Before You Click

Handing an AI your inbox is a bigger ask than handing it a PDF. From Anthropic’s docs:

• Claude looks only when you ask. No question, no peeking.

• Your permissions are the ceiling. A file you can’t open, it can’t open either.

• Claude cannot send email. It drafts. You send. Google’s permission screen lists “send” access; Anthropic has it disabled on their side.

Part 2: Calendar (Steps 1–3)

Step 1. Click + at the left of the message box. Hover Connectors.

One row below Skills. One toggle per app.

Step 2. Toggle Google Calendar. Pick your Google account, read the permissions, approve.

Google’s sign-in and permissions screen for the Calendar connector This is the moment you grant access. Read it.

Step 3. Ask:

What does my Thursday look like? Where's my longest free block this week?

Your week, answered instead of squinted at.

Calendar is two-way. Try “Find 30 minutes with me and Alex next week, afternoons only, and put a hold on it.” It creates, moves, deletes, and RSVPs.

Part 3: Gmail (Steps 4–5)

Step 4. + → Connectors, toggle Gmail, approve.

Step 5. Search like you talk:

Find the thread where we discussed the March invoice. What did we

agree on, and do I owe anyone a reply?

Found and summarized. Note the “Used Gmail integration” receipt.

Then: “Draft the reply. Keep it short.” The draft comes back as a card with Copy and Send via Gmail. Nothing leaves until you click. Add /my-voice (issue 5) and it sounds like you.

Claude has no send button of its own.

One limit: Claude can see attachments, not open them. Download, upload to the chat. Last week’s skill.

Part 4: Drive (Step 6)

Step 6. Toggle Google Drive, approve, ask:

Find my doc about the Q3 planning offsite and pull out the decisions we made.

First reach into a connector, Claude asks: Deny, Always allow, or Allow once. Pick Allow once while you’re learning. Upgrade a connector to Always allow when it has earned it.

Every reach into your stuff gets a yes from you first.

Live version of the doc, yesterday’s edits included.

Docs, Sheets, Slides, PDFs, Office files all work. Know the exact file? + → Add from Google Drive, paste the URL.

Part 5: The Combo Move (Step 7)

Each connector alone is a convenience. Together, they’re the issue.

Step 7. Tomorrow morning, before you open a tab:

Look at my calendar for today. For each meeting, search my email for the latest thread with those attendees and tell me where things

stand. Flag anything I owe someone a reply on, and draft the most

urgent one for my review. End with a brief I can read in one minute.

Calendar, a handful of Gmail searches, cross-referenced, one page. That’s the screenshot at the top.

Part 6: What Else Plugs In (Step 8)

Step 8. + → Connectors → Manage connectors. Settings opens to Customize → Connectors. Click Add → Browse connectors.

Every connected app in one table. The off switches live here too.

2,079 when I looked. You need three.

Don’t connect everything tonight. Each one is an access grant. Add a connector the week a real task needs it.

The Exits

• Disconnect: Settings → Connectors → the app → Disconnect.

• From Google’s side: myaccount.google.com/connections.

• Work account refuses to sign in? IT locked third-party apps. Next week’s topic. Use your personal account today.

• Privacy: Anthropic says it doesn’t train on your connector data. Your issue-4 chat settings still cover what lands in the conversation.

• Incognito chats have no connectors. By design. Use a normal chat.

One Prompt to Steal

Sunday-evening version of the combo move:

Look at my calendar for the coming week and search my email for anything still unresolved from last week. Give me: the three things

next week actually hinges on, anything I've left someone waiting on,

and one meeting I could probably cancel. Be blunt about the last one.

Claude has no politeness stake in your recurring meetings. Its cancel nominations are bolder than yours.

Try It (Right Now): The 20-Minute Wire-Up

1. Connect Calendar. Ask about your week. Have it place one hold.

2. Connect Gmail. Find one buried thread. Draft the reply you owe. Send it yourself.

3. Connect Drive. Fetch one specific doc.

4. Tomorrow morning, run the combo prompt before you open your inbox.

5. Open Settings → Connectors once. Know where the off switches are.

Do those five and Claude is wired into the places your week actually happens.

Next Week

Today you handed Claude your inbox. On a personal account, that’s your call. At work there’s IT, legal, and a fair question about where company data goes. Next week: what’s safe to paste, what never is, and the team-setup basics nobody explains. Forward it to the colleague who’s “not allowed to use AI.” Use Claude at Work Without Getting Fired. See you then.

P.S. Which app do you want wired into Claude that isn’t in the directory yet? Reply and tell me. My money says half of you answer WhatsApp.

Much love ✌️ MJ

🔵 Building. Builds on issues 1 through 6. If your real work lives in files and your Claude chats live in typing, this is the issue that connects them.

New here, or missed one? This is part 7 of a series where every issue stands on the one before it. Catch up in ten minutes:

1 · Do Real Work With Claude in 5 Minutes. Set it up, get your first useful answer. 2 · 3 Small Steps. Save Hours Prompting Claude. Aim, steer, teach. 3 · Stop Overpaying for Claude. Free, Pro, or Max without wasting a dollar. 4 · Claude That Remembers You. Teach it who you are, once. 5 · Claude to Sound Like You. Your voice, saved as a skill. 6 · Stop Sounding Like AI. The six tells and the banned list.

In: a 34-page PDF you were never going to read. Out: the one page you actually needed.

Reading a 34-page contract is an afternoon. Writing the summary off it is the evening. This week, both are a ten-minute job.

You know the folder. The PDF someone sent you “for review.” The spreadsheet with nine tabs answering a question you’ve stopped asking. The whiteboard photo from May, saved forever, read never.

So far this series has worked with what you type. Today we talk about Claude working with what you already have, both directions: it reads your files, and it writes new ones back.

The whole issue up front: flip one switch, upload three files you’ve been dodging (a contract, a spreadsheet, a whiteboard photo), and get three finished pieces of work back, each a real Word doc, Excel sheet, or PDF you can download. Two hours of grunt-work, done in ten. Every plan, Free included. Open Claude in a second tab and grab a file you’ve been avoiding. The avoidance is the point.

Last Week’s Homework

Last week you ran the 10-minute scrub. Here’s the payoff you didn’t see coming: everything Claude writes today, the briefs, the summaries, the doc built from your spreadsheet, comes out through that voice skill. You tuned the instrument. Today it gets real sheet music.

And keep the tells coming from last week’s P.S. The em dash pileup still leads as everyone’s close-the-tab signal, but “It’s not just X” is gaining.

Part 1: Flip On “File Creation”

One switch first. It’s the one that turns “Claude describes your spreadsheet” into “Claude opens it and does the math.”

Step 1. Turn on file creation. Click your initials, bottom-left, then Settings → Capabilities, and scroll to Code execution and file creation. Flip on the toggle marked Cloud code execution and file creation. Same screen where you switched memory on in issue 4.

Same settings page as issue 4, one toggle down. This is the one that makes Claude hand you files back.

It does two jobs: lets Claude run real calculations on your data instead of eyeballing it, and lets it hand back real files (.docx, .xlsx, .pptx, .pdf). Every plan, Free included. Most accounts have it on already, but thirty seconds to check beats a confusing afternoon.

Part 2: What Fits Through the Door

The door is wider than most people think:

• Documents: PDF, Word, TXT, CSV, Excel, and most text formats in the wild.

• Images: photos and screenshots (JPEG, PNG, GIF, WebP). Claude reads them like a person: charts, handwriting, error messages, the whiteboard.

• How they get in: the + button → Add files or photos, or drag the file onto the window and drop it. Pasting a screenshot works too.

• The limits: 20 files per chat, and PDFs under ~100 pages get the full read, text and visuals both. Enough for almost anything you’d hand a colleague.

The + menu. “Add files or photos” up top, or skip the menu entirely and drop the file on the window.

One rule from issue 1 still holds: don’t upload anything you wouldn’t hand to a company. (What’s safe to paste at work is its own issue, coming in issue 9.)

Part 3: Upload Long PDFs

Step 2. Upload it, then brief it like a person. Drag the PDF in. Here’s the move most people miss: don’t ask for a summary. Nobody wants a summary. You want the document handled for your situation. So brief it like a sharp assistant who just walked in:

I have to decide whether to renew this contract by Friday. Read it and
give me: the terms that changed from what you'd consider standard, the
three clauses I should push back on, and any dates or numbers I can't
afford to miss. One page, plain language.

Swap in your life: “I’m presenting this to my boss Monday, pull what she’ll ask about.” “This is my lease, what am I agreeing to in plain English?” That’s issue 2 again: the file is the context, your situation is the aim.

The file is the context. Your Friday deadline is the aim. “Summarize this” wastes the whole setup.

Step 3. Interrogate it. The first reply is the start, not the end. Push back like the document’s a witness: “Where does it say that? Quote the exact line.” “What’s missing that a contract like this should have?” “Explain clause 7 like I’m not a lawyer.” Claude holds the whole file in its head for the rest of the chat, so every follow-up is free.

That “quote the exact line” isn’t decoration. It’s your verification habit, and it comes back before the end.

Part 4: The Nine-Tab Spreadsheet (Steps 4–5)

With code execution on (from Step 1) Claude doesn’t skim your spreadsheet and vibe. It writes and runs real code against your real numbers.

Step 4. Upload the spreadsheet and ask what you’d ask an analyst. Not “analyze this.” A real question:

This is our order export for the last six months. Which products are
growing, which are quietly dying, and is there anything in here that
should worry me? Show your work.

In the reply, Claude opens the tabs, runs the numbers, and answers with figures it computed, not guessed. “Show your work” lets you see the route and catch a wrong assumption early.

Nine tabs in, one answer out. The figures are computed, and “show your work” lets you check the route it took.

Step 5. Ask for the deliverable, not the conversation. The answer doesn’t have to stay in the chat. Say:

Turn that into a one-page summary I can send my team: headline
findings up top, one chart, the numbers behind it in a small table.
Make it a Word doc.

Claude builds the actual .docx and hands it over with a download button. Or an Excel file with working formulas. Or a PDF. You asked a spreadsheet a question and got back a document. That’s the whole loop.

The download button is the moment. Not advice about a document. The document.

Part 5: The Whiteboard Photo You Took But Have Never Used (Step 6)

Step 6. Upload the photo and give the scribbles a job. The whiteboard from May. The napkin sketch. The slide someone flashed for four seconds on a call. Upload it and give it a destination:

This is the whiteboard from our planning meeting. Type up everything
on it, then organize it into: decisions made, open questions, and
action items with owners if you can read them. Flag anything you
can't make out instead of guessing.

Claude reads handwriting better than most people read their own. And that last line matters: “flag what you can’t read” is what separates a transcript you trust from one you have to check against the photo, which defeats the point.

May’s whiteboard, finally read. The “flag what you can’t make out” line is what makes it trustworthy.

Then chain it, now that you know Part 4’s trick: “Turn the action items into a one-page project plan as a Word doc.” Photo in, document out, voice skill on top. Three issues just shook hands in one chat.

Part 6: Keep Editing (Step 7)

Step 7. Treat the file like a draft, not a delivery. Version one is a strong first draft. The magic is that fixing it costs one sentence: “Cut section two in half.” “Make it a bar chart, add March.” “Put the risks first.” Claude edits the file and hands you a fresh download every time.

So don’t download version one and start fixing it in Word out of reflex. Edit by talking until it’s done, then download once. The moment you leave the chat, editing is your job again.

Revisions cost one sentence each. Download last, not first.

Files you’ll reuse belong in a Project. One-offs go in any chat. But the price list, the product spec, the brand guidelines: drop those on a Project’s file shelf (issue 4) and every chat there starts with them already read. Upload once, never re-attach.

One Prompt to Steal

For the file you don’t even know what to do with yet. My intake prompt, before any real request:

Before you do anything with this file, give me:
1. What this document actually is, in two sentences.
2. The five facts in it that matter most.
3. What's missing, outdated, or surprising.
4. The three questions I should be asking about it.
Then stop and let me pick where to go next.

That last line matters. It stops Claude sprinting off to write a report you didn’t ask for, and “three questions” keeps surfacing the thing you actually needed to know. I run it on every contract, every “FYI” report, every spreadsheet I inherit.

The intake pass. Question 4 is where it earns its keep: the questions you didn’t know to ask.

Read the Numbers Twice

One honest warning. Handing files to an AI has a single real failure mode: trusting the output like a calculator.

Claude misreads, rarely but confidently. A smudged scan, a merged cell, a table that wraps wrong, and a number comes out incorrect with a straight face. The fix is the Part 3 habit: for anything that matters, ask “quote the exact line” or “which cells did you use?” Thirty seconds turns “probably right” into “checked.”

Scanned PDFs are photos in a trench coat. A scan of paper reads like an image: usually fine, sometimes garbled. Clean digital PDFs read near-perfectly. Know which you’re holding before you bet a deadline on it.

Big files spend your usage faster. Every page is context Claude holds, and file creation computes on top. On Free and Pro that’s fewer messages per session (issue 3 explains the meter). Upload the chapter you need, not the whole book, and start fresh chats when you switch topics.

Claude sees only what you hand it. One report doesn’t give it your hard drive, your email, or last month’s version. Reassuring for privacy, limiting for work: every file still arrives by hand. Closing that gap is next week.

Try It (Right Now): The File Triathlon

1. Check the toggle. Settings → Capabilities → Code execution and file creation. (Step 1)

2. Feed it the PDF you’ve been avoiding with a brief about your actual situation, then interrogate it. (Steps 2–3)

3. Feed it a spreadsheet and ask a real question, then ask for the answer as a Word doc. (Steps 4–5)

4. Feed it a photo of handwriting or a whiteboard, with “flag what you can’t read.” (Step 6)

5. Edit one output by talking, then download the final version. (Step 7)

6. Run the intake prompt on the next file anyone sends you.

Do those six and the folder of good intentions stops being a folder of guilt. It’s raw material now.

Next Week

Today you fed Claude files by hand, one drag at a time. Next week we lose the hand. Connectors wire Claude into your email, calendar, and drive, so instead of uploading the meeting notes you say “find the meeting notes” and it goes and looks. That’s the jump from feeding Claude to giving it a seat at your desk. Plug Claude Into Your Life. See you then.

P.S. What’s the oldest file in your “I’ll deal with it later” folder, and what happened when you finally fed it to Claude this week? Reply and tell me. Best redemption story gets a shout-out next issue. Mine was a 2024 lease renewal I “reviewed” eleven months after signing it.

Much love ✌️ MJ

🔵 Building. Builds on issue 5. If you’ve ever hovered over send wondering “can they tell?”, this is the issue that lets you stop wondering.

New here, or missed one? This is part 6 of a series where every issue stands on the one before it. Catch up in ten minutes:

1 · Do Real Work With Claude in 5 Minutes. Set it up, get your first useful answer. 2 · 3 Small Steps. Save Hours Prompting Claude. Aim, steer, teach. 3 · Stop Overpaying for Claude (publishing soon). Free, Pro, or Max without wasting a dollar. 4 · Claude That Remembers You (publishing soon). Teach it who you are, once. 5 · Claude to Sound Like You (publishing soon). Your voice, saved as a skill.

Same request, same Claude. The left one had nothing. The right one had the rules you’re about to steal.

You know the feeling. Two lines into a LinkedIn post, you close it. Nobody taught you the rules it broke. You just smelled robot.

Your readers have the same radar. So here’s the uncomfortable part: even with the voice skill you built last week, Claude’s drafts still slip into robot mode. An em dash pileup here, a “this isn’t just a launch, it’s a statement” there, and people quietly stop reading.

Good news: only six tells do most of the damage. Learn them once and you’ll spot them in seconds. Even better, you can teach Claude to delete them before a draft ever reaches you.

The answer up front, as always: learn the six tells, steal my banned-phrases list (the one this newsletter runs on), and paste it into last week’s my-voice skill. Drafts come out pre-scrubbed. Ten minutes. Works on every plan, Free included.

Last Week’s Homework

Last week you turned your best writing into a voice guide and saved it as a /my-voice skill. If you ran the blind test, you saw it work: drafts started sounding like you.

Today we add the other half. The voice guide covers what you sound like. It says nothing about the robot habits, and those leak into anyone’s voice. So the same skill gets a second section: the do-not list.

Last week’s P.S. asked which phrase Claude keeps putting in your mouth. Keep those coming. Every one belongs on the list you’re about to build.

Why the Drafts Sound Like That

Quick detour. Once you know why the tells happen, you’ll catch ones I never list.

Claude picks the most likely next word, and “most likely” means “what everyone writes.” Left alone, the writing drifts to the middle: safe and beige. When it’s not sure a point landed, it pads, and the padding is where the “pivotal moments” and the little summary bows come from.

That’s the house style of every AI model out there. Your reader has seen it ten thousand times. This time your name is on it.

The Six Tells

Learn these once and you’ll never unsee them, in Claude’s drafts or in your feed.

1. The em dash pileup

The most famous tell, fair or not. One dash is punctuation. Four in a paragraph is a signature:

“The launch went well — better than expected, honestly — though the real story is what came after — the retention numbers.”

Nobody types that. A machine does, constantly. The fix is boring on purpose: periods, commas, and parentheses do the same work quieter. (House rule around here: this newsletter bans em dashes outright.)

Straight from Claude, unprompted: the double-dash aside, mid-sentence. Nobody punctuates like this by hand.

2. “It’s not X. It’s Y.”

The fake-deep contrast:

“This isn’t just a product update. It’s a statement of intent.” “Success isn’t about working harder. It’s about working smarter.”

The shape performs insight without containing any. Delete the first half and say the point straight: “This update shows we’re serious about enterprise.” If nothing useful is left once the setup is gone, there was no point to begin with.

Insight-shaped, insight-free. Delete the setup and see what’s left.

3. Everything comes in threes

“Faster, cheaper, and more reliable.” “Clarity, confidence, and control.” Triples feel complete, so the machine reaches for them even when only one item is true. Real writing is lumpy: one reason here, four over there. When you spot a matched set of three, ask which item is doing the work and cut the passengers.

4. Brochure words

Delve, leverage, seamless, robust, elevate, unlock, empower, game-changer, transformative, journey, landscape. The test is simple and brutal: would you say it across a table? “We can leverage this” no. “We can use this” yes.

Same family, different disguise: inflation. Not every update is “exciting,” not every decision is “crucial,” and almost nothing is “a pivotal moment.” Say the flat true thing and let the reader decide how big it is.

5. Vague where a fact should be

The biggest tell, and the one that survives careless editing. Compare:

“The team faced significant challenges this quarter.” “The team missed payroll twice this quarter.”

Machines abstract because they don’t know your specifics. You do, and that’s your entire advantage, so spend it. A draft with no number, name, or date anywhere reads like a form letter, because it is one. (This is also issue 2 paying off again: specifics can’t show up in the draft if they never went into the brief.)

Same update, same facts available. Only one sounds like a person who was in the room.

6. The wind-up and the bow

Openings that clear their throat: “In today’s fast-paced world,” “I hope this email finds you well.” Endings that restate the middle: “In conclusion,” “Ultimately, what this means is.” Plus the eager-assistant residue: “I hope this helps!”

Start where the point starts. Stop at the last point. The reader can handle the rest.

That’s all six. Most drafts carry two or three, and once you know the list, spotting them takes seconds.

All six on one draft. This is the ten-second scan you’ll be running by Friday without noticing.

The Banned List (Steal This Whole Block)

Here’s the deliverable: my working rules file, boiled down to a block that fits in a skill. This issue was edited against the full version of it.

## The anti-AI pass Before showing me any draft, edit it against these rules:

- No em dashes, ever. Use a period, comma, colon, or parentheses.

- Never write "It's not X, it's Y" or any sentence that rejects one

  idea to assert another. State the point directly instead.

- Banned words: delve, leverage, seamless, robust, elevate, unlock,

  empower, harness, streamline, game-changer, transformative,

  tapestry, landscape, journey (unless someone is actually traveling).

- Banned phrases: "I hope this finds you well", "in today's

  fast-paced world", "let's dive in", "it's important to note",

  "in conclusion", "I hope this helps".

- Don't force lists of three. Use the real number of points.

- Replace every vague claim with a number, name, date, or example.

  If you don't have one, ask me for it instead of writing filler.

- No warm-up intros and no summary endings. Start at the point,

  stop at the last one.

- Vary sentence length. If three sentences in a row have the same

  shape, break one.

That’s a starter kit, tuned to the tells above. It grows from here: every time you catch Claude putting a phrase in your mouth, add it to the bottom. Your list a month from now will be personal in a way mine can’t be.

Bolt It Onto Your Voice Skill (2 Clicks and a Paste)

Short click-along this week. Open Claude in a second tab.

Step 1. Open your skill for editing. In any chat, click the + in the message box, then Skills → Manage skills. Your Skills page opens. Click your my-voice row from last week; it opens in a read-only view showing your voice guide. Now click the ⋮ menu in the top-right corner and pick Edit. A form titled Edit skill instructions opens, the same three fields you filled when you created it.

Same page as last week. Your my-voice row is about to get smarter.

Step 2. Paste the block at the bottom of the Instructions box and hit Save. Scroll to the end of your voice guide, paste the anti-AI pass under it, Save. A little “Saved changes to my-voice” toast confirms it. Done. Your skill now has two halves: the voice guide (what you sound like) and the banned list (what to delete). Every draft that goes through /my-voice, or gets the skill auto-loaded by its description, comes out pre-scrubbed.

⋮ → Edit, paste at the bottom, Save. One skill, two halves: voice on top, banned list underneath.

Skipped issue 5? You can still have this today. Same Skills page, click Add → Write skill instructions, and fill the form: name no-ai-tells, description Edit drafts to remove AI-sounding writing. Use whenever writing or rewriting anything for me., then paste the block as the instructions and hit Create. That’s the banned list as a standalone skill. It stacks fine with a voice skill whenever you build one.

The no-voice-skill-yet version. Ten minutes, still worth it.

One Prompt to Steal

The skill covers what Claude writes for you. This one is for everything else: the report a colleague “wrote,” last year’s website copy, your own writing on an off day. Paste it above any text:

Run an anti-AI pass on the text below. Flag every tell you find: em dashes, "not X but Y" contrasts, forced lists of three, buzzwords,

vague claims with no specifics, warm-up intros, summary endings.

Show me the flagged list first, then the cleaned rewrite. Keep the

meaning and the voice. Change nothing that already sounds human.

[paste the text]

Don’t skip the flag list to get to the rewrite. Watch which tells show up in your own writing three drafts in a row. Those are yours. Knowing them by name is how you stop typing them.

The audit. The rewrite is useful. The flag list is the education.

Don’t Overcorrect

The honesty section, because this issue has a failure mode.

The tells are innocent one at a time. Em dashes predate the light bulb, and your favorite novelist uses them beautifully. “Delve” is a real word with a real job. What gives a draft away is the pile: dashes plus a triple plus “elevate” plus zero specifics, all in one email. You’re deleting the pile.

Scrubbing too hard is its own tell. Prose that visibly dodges every AI-ism goes stiff, and stiff reads machine-made too. If a fix makes the sentence worse, keep the original. The test never changes: read it aloud. If it sounds like you on a decent day, send it.

Claude will still slip. The skill raises the floor a lot; it doesn’t retire you as editor. Long drafts drift back toward house style, so keep a ten-second scan as the last thing before send. Cheap insurance for the thing that has your name on it.

Try It (Right Now): The 10-Minute Scrub

1. Generate a specimen. Fresh chat, no skill, ask: “Write 200 words on why meetings run long.” A plain request like that produces tells on demand.

2. Hunt. Count the dashes, the triples, the brochure words. My over/under is four.

3. Paste the banned list into my-voice (or create no-ai-tells if you skipped issue 5).

4. Rerun the exact same request with the skill on. Read both versions out loud, back to back. That’s the whole issue in one A/B test.

5. Add one personal rule to the list: the phrase you’d never say that keeps showing up anyway.

The A/B that settles it. Same request, ten minutes of setup apart.

Do that and the next thing you send won’t trip anyone’s robot alarm, including yours.

Next Week

So far the series has Claude working with what you type. Next week it starts working with what you already have: the PDF you’ve been avoiding, the spreadsheet with nine tabs, the photo of the whiteboard. Upload them and get back finished work you keep editing. Feed Claude Your Files. See you then.

P.S. Which tell is your instant close-the-tab signal? Mine is the em dash pileup, though “It’s not just X” is gaining fast. Reply and tell me. The banned list grows one reader at a time.

Much love ✌️ MJ

🔵 Building. Builds on issues 1 through 4. For anyone who rewrites every draft Claude hands them before it’s fit to send.

New here, or missed one? This is part 5 of a series where every issue stands on the one before it. Catch up in ten minutes:

1 · Do Real Work With Claude in 5 Minutes. Set it up, get your first useful answer. 2 · 3 Small Steps. Save Hours Prompting Claude. Aim, steer, teach. 3 · Stop Overpaying for Claude (publishing soon). Free, Pro, or Max without wasting a dollar. 4 · Claude That Remembers You (publishing soon). Teach it who you are, once.

Same request, same Claude. The left one is fine. The right one you could send without touching.

Ask Claude to draft a quick note to your team. “I hope this message finds you well.” “I wanted to reach out regarding.” “Please don’t hesitate to let me know if you have any questions.” Professional, sure. Also three sentences you have never once said to people you see every day.

So you fix it. Cut the greeting, unstiffen the middle, delete “furthermore.” Eight edits later it sounds like you, which means the draft saved you five minutes and the rewrite cost you six.

Last week you gave Claude a memory, so it knows who you are. Today it learns how you sound: how you open, how long your sentences run, the words you’d never let past your fingers.

Here’s the whole play: collect five samples of your real writing, have Claude turn them into a voice guide, then save that guide as a Skill you can drop into any chat with one keystroke. Twenty minutes. Works on every plan, Free included. Open Claude in a second tab and click along.

Last Week’s Homework

Your task was the memory interview: seven questions, and Claude wrote up who you are. Memory answers who’s asking. Today’s guide answers how they’d say it. Skipped it? Today still stands on its own, but go back for issue 4 when you’ve got ten minutes.

Part 1: Pull Your Receipts (10 Minutes, No Clicks)

Claude can’t imitate a voice it’s never heard. So before we touch a single menu, collect 3 to 5 samples of your real writing into one document or note. This is the raw material for everything else.

What counts as a good sample:

• You wrote it alone. No committee edits, no template. The all-hands doc that legal smoothed over is not your voice.

• You’d send it again. Pick writing you were quietly pleased with: the email that landed, the post that got replies, the intro paragraph you didn’t hate.

• It’s long enough to have rhythm. A paragraph minimum. One-line “sounds good!” replies teach nothing.

• Mix the settings. One work email, one longer doc, one casual message. The mix is what shows your range.

Five strong samples beat twenty lukewarm ones. Show it your best swing.

Part 2: Make Claude Write Your Voice Guide (Steps 1–2)

Step 1. Paste the prompt, then the samples. Open a new chat and send this with your samples pasted at the bottom:

I want you to learn to write like me. Below are samples of my real writing. Study them and write my voice guide: how I open, my sentence

rhythm, how I use punctuation, words and phrases I reach for, words I

would clearly never use, how I handle humor, and how I close. Write

the guide as direct instructions to a ghostwriter ("Open with the

point. Keep most sentences under 15 words."), not as observations

about me. End with a list of 5 phrases that sound like AI or corporate

filler that I never use, so we can ban them outright.

Here are my samples: [paste your 3 to 5 samples]

What comes back is a page of instructions: your openers, your pacing, your favorite moves, your never-words. Reading it is a little unsettling, in the way hearing a recording of your own voice is. That’s the sign it’s working.

Your voice, written down as instructions. Slightly creepy. Extremely useful.

Step 2. Correct it out loud. The first guide is a strong sketch with a few wrong lines. Read it and talk back: “I never open with a question, cut that.” “The exclamation points are real, keep them, but cap it at one per message.” “Add that I write numbers as digits.” Claude revises the guide on the spot. Two or three rounds of this and the guide stops describing someone almost like you and starts describing you.

When you’re happy, ask for the final version in one clean block. Copy it. That’s your master file, and in the next part we make it permanent.

Part 3: Save It as a Skill (Steps 3–5)

Right now that voice guide lives in one chat, and chats scroll away. The fix is a Skill: a saved set of instructions Claude can load into any conversation. You write it once, it’s available forever, on every plan including Free.

Step 3. Open your Skills page. In any chat, click the + on the left end of the message box, go to Skills, then click Manage skills. Settings opens to a page called Skills with a table of what you’ve got installed. (You can also get here through Settings directly: it’s under the Customize section in the sidebar.)

The + menu in the message box. Skills, then Manage skills.

Your Skills page. Today it gains its most useful row.

Step 4. Click Add, then “Write skill instructions.” Top right of the Skills page there’s an Add button. Click it and pick Write skill instructions from the little menu. (The other two options, “Create with Claude” and “Upload a skill,” are real and we’ll meet them in a later issue. Today the paste-it-yourself route is the fastest.)

Three ways in. “Write skill instructions” is ours: we already have the instructions.

Step 5. Fill the three fields and hit Create. A form opens with three boxes:

• Skill name: my-voice. Short, no spaces.

• Description: this field matters more than it looks. Claude reads it to decide when to pull this skill in on its own. Write: Write in my voice. Use whenever drafting, rewriting, or editing emails, posts, or documents for me.

• Instructions: paste your entire voice guide from Part 2.

Click Create. Your voice now lives in Claude permanently, one row in that table, editable any time.

Name, description, instructions. The description is the trigger: it tells Claude when to use this on its own.

Part 4: Use It Everywhere (Steps 6–7)

Step 6. Call it with a slash. In any chat, type / in the message box and your skills pop up, each with its description. Pick my-voice (or click + → Skills → my-voice, same thing).

Type / and your skills appear. That description you wrote? It rides along as the tooltip.

A little /my-voice token drops into your message. Finish the sentence, “draft a short email to my team announcing that the weekly sync is moving,” and send. The draft comes back in your voice, opener to sign-off. (Grabbed it by mistake? Backspace deletes the token like any other word.)

One keystroke. The token loads your whole voice guide into the request.

Step 7. Then notice when you don’t have to. Remember that description field? Because it says “use whenever drafting emails, posts, or documents for me,” Claude will often load the skill unprompted the moment you ask it to write something as you. The slash is for when you want to force it; the description handles the everyday case. Set once, sounds like you by default.

Part 5: The Blind Test (Step 8)

Step 8. Run the same request twice. Open a fresh chat and ask for a draft cold, no skill. Then ask again in another chat with /my-voice. Put them side by side.

The plain one reads like a very polite stranger. The skill one should read like you on a decent morning. Anywhere it slips, and it will somewhere, go back to Settings → Skills, click my-voice, open the ⋮ menu, choose Edit, and tighten the line: “Never say ‘thrilled.’” “Stop opening with ‘Quick question.’” Hit Save. The skill is a living document. Every miss you correct makes next week’s drafts better, and thirty seconds of editing beats re-teaching it in every chat.

The skill run in full: one token, one request, and a draft that ships as is. If yours doesn’t sound like you yet, edit the skill, not the draft.

Your Three Layers, Sorted

You’ve now built three separate pieces, and each has one job:

• Memory (issue 4): who you are. Role, projects, standing preferences.

• Your voice skill (today): how you write. Loaded on demand or auto-triggered.

• Project instructions (issue 4, Part 7): the assignment. Per-job briefs like “you’re my editor, push back on flab.”

They stack. A chat inside your Newsletter project with /my-voice active knows the job, knows you, and sounds like you. That’s the whole personalization picture, and you built all of it in two issues.

One Prompt to Steal

For the draft you already wrote but aren’t sure about. Paste your text under this:

Rewrite this in my voice using my my-voice skill. Then list the three

phrases in my original that most sound like they were written by AI

or a committee, and what you replaced them with.

You get a cleaner draft plus a mirror: the exact spots where your own writing drifts into template-speak. Half the value is the rewrite. The other half is watching the same three tells show up every time, until you stop typing them.

A Couple of Gotchas

Expect a strong impression, not a clone. A good voice skill gets you 80% of the way, which turns “rewrite the whole thing” into “change four words.” You still read before you send. You just stop dreading it.

The guide is only as good as the samples. If the drafts feel off, the fix is almost always upstream: swap in stronger samples and regenerate the guide before you start patching rules.

Rules beat adjectives. “Warm but professional” steers nothing. “Sentences under 15 words, contractions always, one exclamation point max, never ‘circle back’” is a guide a ghostwriter could actually follow. When you edit the skill, add rules like the second kind.

If your + menu says “Use style” instead of Skills. You’re on the older surface for the same idea; Anthropic is folding Styles into Skills, and the rollout reaches accounts at different times. The moves translate: create a custom style from your writing samples there, and migrate when your menu changes.

Try It (Right Now): The 20-Minute Voice Transplant

The whole issue as a checklist:

1. Collect 3 to 5 samples of writing you’re proud of into one note. (Part 1)

2. Run the voice-guide prompt with your samples pasted in. (Step 1)

3. Correct the guide out loud until reading it feels like being caught. (Step 2)

4. + → Skills → Manage skills → Add → Write skill instructions. (Steps 3–4)

5. Name it my-voice, give it the “use whenever drafting” description, paste the guide, Create. (Step 5)

6. Draft something real with /my-voice and send the result to an actual human. (Step 6)

7. Run the blind test and tighten one rule based on what slipped. (Step 8)

Do those seven and the next draft Claude hands you will be one you barely touch.

Next Week

Your voice skill fixes what Claude should sound like. Next week we go after the opposite problem: the tells that make writing sound like AI no matter whose voice it’s wearing. The em dash addiction. “Delve.” The tidy little summary paragraph nobody asked for. I’ll hand you my actual banned-phrases list, the one I run every issue of this newsletter through, and you’ll bolt it straight onto the skill you built today. Stop Sounding Like AI. See you then.

P.S. What’s the one phrase Claude keeps putting in your mouth that you would never actually say? Hit reply and tell me. I’m collecting them for next week’s issue, and the current front-runner is “I hope this finds you well,” which has never once found anyone well.

Much love ✌️ MJ

🟢 Beginner. Builds on issues 1 through 3. If you keep re-typing the same background every time you open a chat, this one hands you those minutes back.

New here, or missed one? This is part 4 of a series where every issue stands on the one before it. Catch up in ten minutes:

1 · Do Real Work With Claude in 5 Minutes. Set it up, get your first useful answer. 2 · 3 Small Steps. Save Hours Prompting Claude. Aim, steer, teach. 3 · Stop Overpaying for Claude (publishing soon). Free, Pro, or Max without wasting a dollar.

This little panel is Claude’s memory of you. Today we fill it on purpose.

Every new chat, you re-explain yourself. Your job. Your project. The tone you like. That you want the answer first and the caveats after. You typed all of that last week, and the chat you opened this morning has no idea. It starts from scratch every time.

That ends today. Claude has a memory now. It’s on for everyone, free plans included, and once you set it up it carries what it knows into every new chat. You stop introducing yourself to it every morning.

This one’s a click-along, same as last week, but tighter. I’ll walk you through the clicks to set memory up properly, in order, one at a time. Open Claude in a second tab right now and do each step on your own account as we go. Fifteen minutes from now Claude will know your role, your goals, and how you like to be talked to, and you’ll never type that intro paragraph again.

The answer up front, because I won’t bury it: memory is already on and already learning you. The work is spending fifteen minutes teaching it the right things on purpose instead of leaving it to guess. Let’s go do that, step by step.

Last Week’s Homework

Last week you opened your plan screen, checked your meter, and made an honest call: Free, Pro, or Max.

Whatever you landed on, memory works the same. This is one of those free features issue 3 promised was real, not a stripped-down demo. So you don’t need to have upgraded anything to do today’s work. If you’re still on Free, you’re fully invited. (The one nearby thing that is paid-only is searching your old chats by keyword. Memory itself, the part that remembers you, costs nothing.)

Part 1: Find the Switch (2 clicks)

Before you teach memory anything, you need to know where it lives. Two clicks.

Step 1. Open your account menu. Look at the bottom-left corner of Claude. Your name and a small icon sit there. Click it. A little menu pops up. This is the same door from last week, the one that leads to everything about your account.

Bottom-left corner. Click your name to open this menu, then click Settings.

Step 2. Go to Capabilities. In that menu, click Settings. When Settings opens, look down the left sidebar for Capabilities and click it. This is the page that controls what Claude can and can’t do, memory included.

Settings → Capabilities. Everything about memory lives on this one page.

Part 2: Confirm Memory Is On (1 click)

Step 3. Find the memory switch and check it’s on. On the Capabilities page, look under the Memory heading. You’ll see a switch labeled Generate memory from chat history, described as “Allow Claude to remember relevant context from your chats. This setting controls memory for both chats and projects.” That’s the engine. It’s almost certainly already on, but look with your own eyes. If it’s off, click it on.

(There’s a second switch just above it, Search and reference chats. Different thing: that one lets Claude dig up specific old conversations, and it’s the paid-plan perk we mentioned. The one you care about today is Generate memory from chat history, and that’s on for everyone.)

When memory’s on, Claude builds a short summary of who you are from your chats and reuses it in every new conversation. When it’s off, every chat starts from zero.

The switch labeled “Generate memory from chat history” is the engine. If it’s already blue, leave it.

Part 3: See What It Already Knows (1 click)

Step 4. Open your memory. Right under the switch is a row that reads View and manage memory (with a note like “Updated 18 hours ago”). Click it. A panel titled Manage memory slides open.

Up top it tells you the deal: “Here’s what Claude remembers about you. This summary is regenerated each night and does not include projects, which have their own specific memory.” Below that is the summary itself, sorted into plain-English sections, usually Work context, Personal context, and Top of mind. If you’ve used Claude for a few weeks, it’s probably already picked up a thing or two: your job, a project you keep mentioning, how you like to write. Read it now. This is the exact context Claude silently pastes into the top of every new chat you start.

Everything Claude knows about you, in plain sentences, sorted into sections. Read it before you change it.

Two things to notice. First, it’s a short summary, not a transcript. Claude isn’t hoarding your conversations word for word, it’s keeping a tidy few paragraphs of what matters, and it rewrites that summary fresh every night. Second, it’s yours to change. Nothing in here is locked. That’s the whole next part.

Part 4: Teach It On Purpose

Here’s where you stop letting Claude guess and start telling it. It’s the most useful thing in this whole issue, and almost nobody bothers. Don’t skip it.

Step 5. Say “Remember that…” out loud. Open any chat and type a sentence that starts with “Remember that…” Try this exact shape:

Remember that I run marketing for a 12-person B2B software company, I write in a plain, direct style with no corporate buzzwords, and I

always want the recommendation first and the reasoning after.

Send it. Claude confirms it’s got it and folds that straight into your memory summary. From your very next chat, it knows all of that before you type a word. You just deleted the intro paragraph you’ve been re-writing for weeks.

Say “Remember that…” and it’s saved instantly. Next chat, Claude already knows.

Step 6. Do it for the handful of things that are true every time. One sentence each. Cover these:

• Who you are. Your role, your company or field, what you’re usually trying to get done.

• How you like answers. Short or thorough. Bullet points or prose. Answer-first or context-first. No emoji, or bring the emoji.

• The stuff you’d otherwise re-explain. The product you’re building, the audience you write for, the two clients whose names keep coming up.

Spend three minutes on that and you’re done. You’re telling memory what matters instead of leaving it to guess.

Part 5: Clean It Up (1 click)

Memory rewrites that nightly summary from your chats on its own, so it occasionally picks up something crooked, a one-off you mentioned that it mistook for a standing fact.

Step 7. Prune the individual memories. In the Manage memory panel, click Manage edits. This opens the list of specific things Claude is holding onto, each on its own line: “Prefers to be called by a nickname,” “Never use exclamation points,” “Working on the Q3 launch,” and so on. Each line has a small trash icon. See something wrong or out of date? Click the trash to drop it. Want to wipe your manual edits and start clean? There’s a Clear edits option right there too.

Manage edits shows each remembered item on its own line. Trash the ones that are wrong or stale.

A little cleanup here saves you from Claude assuming you’re still on a project you wrapped months ago.

Part 6: Know Your Privacy Valves (3 controls)

Memory should always feel like something you own, not something happening to you. Three controls do that. Learn where they are even if you don’t use them today.

Step 8. The one-off chat (ghost icon). Some conversations you just don’t want remembered: a draft you’re venting into, a search that isn’t really “you.” When you start a new chat, look for the little ghost icon. Click it and that chat runs incognito. Claude won’t save it to memory or your history. Say your piece, close it, it’s gone.

Click the ghost icon and you’re incognito. The screen says it plainly: nothing here is saved to memory or your history.

Step 9. Pause it, or wipe it. Back on Settings → Capabilities you’ve got the bigger switches:

• Pause is just that memory switch. Flip Generate memory from chat history off and Claude keeps what it already knows but stops using it and stops learning anything new, until you flip it back. Good for a stretch of work you want kept totally separate.

• Wipe lives in Manage edits. Trash individual memories one at a time, or hit Clear edits to reset the whole thing to a clean slate. You’ll rarely want the full wipe, but it’s worth knowing where it is.

The wipe lives in Manage edits: trash items one at a time, or Clear edits to reset to a clean slate.

Part 7: Give Each Big Job Its Own Brain (Projects)

Here’s where memory gets really useful. Everything above is Claude’s memory of you, one shared summary across all your everyday chats. But you don’t want your novel bleeding into your tax questions bleeding into your day job. That’s what Projects are for.

A Project is a dedicated workspace with its own separate memory, walled off from your other chats and from each other.

Step 10. Create a Project. In the left sidebar, find Projects and click New project (or the + next to it). Give it a name, something like “Newsletter” or “Job Search.” That’s it, you’ve got a walled-off workspace.

Projects live in the left sidebar. Each one is its own sealed brain.

Step 11. Give it a standing instruction. Open your new Project and find the instructions field. Type the brief that applies to every chat inside it. For a newsletter Project: “You’re my editor for this newsletter. Match my voice. Push back when a sentence is flabby.” Set once, applies to every conversation in that Project, forever.

One instruction, and every chat in this Project already knows the assignment.

Step 12. Feed it the files this job leans on. In the same Project, add the background docs, the style guide, the spreadsheet, whatever this particular job needs. Every chat in the Project can see them, and none of it clutters your general memory.

Every Project has a Files shelf. Drop in the docs this job leans on, and every chat inside it can see them.

So your “Newsletter” Project remembers your voice and your past issues. Your “Job Search” Project remembers your resume and the roles you’re chasing. Neither one knows about the other, and neither one clutters your general memory. Each is its own labeled folder, already stocked with what that job needs.

We’ll go deep on Projects and files in later issues. For today, just create one, give it a one-line instruction, and notice how much smoother it goes when the chat already knows the assignment.

One Prompt to Steal

If you’d rather not think up your whole profile from scratch, let Claude interview you and write it for you. Paste this into any chat, Free included:

Interview me so you can build a great long-term memory of me. Ask me 6 or 7 quick questions, one at a time, about my job, what I mostly use

you for, the tone and format I like in answers, and anything I'd

otherwise have to re-explain every chat. When we're done, write it up

as a clean "Remember that…" summary I can approve, then save it.

It asks the questions you wouldn’t have thought to ask yourself, then hands you a tidy memory summary to approve. Two minutes, and Claude knows you better than it would after a month of guessing.

Hand it the setup. It interviews you, then writes a clean memory summary to approve.

A Couple of Gotchas

Memory isn’t a filing cabinet. It keeps a short summary of what matters about you, not a perfect transcript of every chat. Don’t rely on it to fetch the exact number you mentioned three weeks ago. For that you want a Project with the actual file in it, which is exactly where we’re headed.

Automatic memory has a slight delay. The stuff it learns on its own refreshes roughly once a day. But anything you say with “Remember that…” lands right away. When it matters, tell it outright instead of hoping it noticed.

Keep it tidy, not big. More isn’t better. A clean paragraph that’s actually true will steer Claude better than a pile of half-relevant facts. Prune the junk now and then.

Try It (Right Now): The 5-Minute Setup

If you clicked along, you’re already done. If you skimmed, here’s the whole thing as a checklist. Do these in order and you’ve set memory up like a pro:

1. Open the switch. Bottom-left → Settings → Capabilities. (Steps 1–2)

2. Confirm Memory is on. Toggle checked. (Step 3)

3. Read what it knows. Click “View and manage memory.” (Step 4)

4. Teach it the essentials. Three “Remember that…” lines: who you are, how you like answers, the thing you always re-explain. This is the move. (Steps 5–6)

5. Tidy it. Manage edits, trash anything crooked or stale. (Step 7)

6. Know your valves. Locate the ghost icon, Pause, and Reset so memory always feels like yours. (Steps 8–9)

7. Spin up one Project. Name it, give it a one-line instruction, drop in a file. (Steps 10–12)

Do those seven and you’ve retired the intro paragraph forever. Tomorrow morning, open a fresh chat and just ask for the thing. Claude already knows who’s asking.

Next Week

Issue 5 takes this one step deeper. Memory means Claude knows who you are. Next we make it sound like you: your phrasing, your rhythm, the words you’d actually use, so what it writes reads like you wrote it and not like a robot doing an impression of you. That’s what turns a draft you have to rewrite into one you can send as is.

See you then.

P.S. What’s the first thing you told Claude to remember? Hit reply and tell me. I collect the good ones, and the sharpest “Remember that…” lines I’ve seen are almost always about format, not facts. Curious what you reach for first.

Much love ✌️ MJ

Free, Pro, Max. The whole decision on one screen.

You signed up free back in issue 1. You got good at asking in issue 2. Now let’s make sure you’re on the right plan, so you’re not overpaying and not stuck behind a wall you’ve already outgrown.

This one works differently from the last two. It’s a guided tour of the actual screens inside Claude, and I want you clicking along with me the whole way. Open Claude in a second tab right now. Every time I say “click here,” go do it on your own account. Ten minutes from now you’ll have sized up your real usage, seen every plan with your own eyes, and made the call with data instead of a guess.

The answer up front, because I won’t bury it: most people should ride Free until it hurts, then jump straight to Pro. Almost nobody needs Max yet. The clicks below are how you find out which of those is you.

Last Week’s Homework

Last week you ran all three steps on one real task: aim it, steer it, teach it.

If you did that a few times, you already have the one thing that decides this whole question: a feel for how often you actually reach for Claude, and how heavy your asks are. That beats any chart I could draw you. We’re just going to go confirm it with the numbers on your own screen.

Open the One Screen This Whole Issue Lives On

Everything today happens in one place, so let’s get you there before anything else.

Look at the bottom-left corner of Claude. Your name and a small icon sit there. Click it, then click Settings, then find Plan (a few accounts label it Billing). That’s the control room. Upgrade, downgrade, switch from monthly to yearly, it all happens on this one screen. Leave it open, we’ll keep coming back to it.

Bottom-left corner. This little menu is the door to everything about your plan.

Find Your Meter (Free Has a Catch)

Before you look at a single price, get honest about how hard you actually lean on Claude. And here’s the catch nobody warns you about: on the free plan, there’s no usage meter to open. Pop into Settings and the sidebar reads General, Account, Privacy, Billing, Capabilities, Connectors, Claude Code. Go looking for “Usage.” It isn’t there.

So on free, your meter is the wall. The only reading you get is the moment Claude stops mid-work and tells you you’re out for a few hours, with your reset time and an Upgrade button sitting right there. How often that message shows up is the whole signal. Once a month? You’re fine. Every couple of days? You’ve outgrown free.

That little gray box is the most useful thing in this whole issue. It’s Claude telling you you’ve hit your tier’s ceiling, and it’s the only buy signal you need. Not a LinkedIn post, not FOMO. The wall.

Quick word on how the limit works, since it trips everyone up. Claude doesn’t give you a fixed number of messages a day. Your usage refills on a rolling window of a few hours, and the paid plans add a weekly ceiling behind that. Three things drain it fastest: long back-and-forths, leaning on the big Opus model instead of everyday Sonnet, and heavy attachments like a 40-page PDF. That’s why nobody can promise “you get 45 messages.” It genuinely depends.

One perk worth knowing now: the day you move to Pro or Max, you unlock a real Settings → Usage page, an actual dashboard with bars and a live reset clock. On free you fly by the wall. On paid you get the gauge.

Walk the Three Plans With Me

Back to the plan screen you left open. You’re looking at three cards. Let’s read each one together and I’ll tell you who it’s actually for.

Free — $0 (the card you’re on right now)

Look at the Free card. It’s not a crippled demo, which is the whole reason issues 1 and 2 never asked you to pay.

What it gives you: the everyday Sonnet model, chat on web, desktop, and phone, file uploads, image reading, web search, extended thinking for the hard questions, memory across chats, and connectors. Most of what made you go “oh, neat” so far.

Where it stops: you run out of room the fastest, there’s no Opus (the big-brain model), and no hand-off features (Cowork and Claude Code) that the back half of this series is built on. When servers are slammed, paid users go first.

Stay here if you use Claude a few times a day, the wall rarely shows up, and nothing above feels like something you’re missing. Don’t upgrade out of guilt.

Free isn’t a stripped-down demo. Real work lives here.

Pro — $20/month (click Upgrade to see it)

Click the Upgrade button under Pro. Don’t buy anything yet. Just look at what the screen lists, so the $20 stops being abstract.

What unlocks: a lot more room before the wall, plus the good stuff. The headline is Opus, the model you call for heavy lifting. Then unlimited Projects (issue 4’s whole subject), Research for those read-fifty-sources briefings, and the two delegation features the series leans on later, Cowork and Claude Code. Priority at busy times rides along.

Jump here when the wall shows up more than once or twice a week, or you want any single one of Opus, Projects, Cowork, or Code. One of them is reason enough.

The catch is that it’s $20 a month. That’s it. Use Claude for real work even a few times a week and it pays for itself the first hour it saves you.

What the $20 buys: Claude Code, Cowork, more models, and a lot more room.

Max — $100/month (5x) or $200/month (20x)

Scroll to the Max card and you’ll see two prices. 5x is five times Pro’s usage for $100. 20x is twenty times for $200. Same features as Pro, plus early access to new stuff and front-of-line priority. Max also splits its ceiling into two weekly limits, one across all models and one just for Sonnet, so heavy users get a saner rhythm.

Max is really two plans. Start at 5x. Only climb to 20x if you cap that out.

For the record, I’m on Max 5x, the $100 one, and I regularly max it out. That’s not a flex, it’s the exact point I’m making: I know it’s the right plan for me because I keep hitting its ceiling. If I coasted along never touching the limit, I’d be lighting $100 on fire every month and I’d drop to Pro tomorrow. The plan you should be on is the one whose wall you actually bump into.

The honest part: if you’re reading a beginner newsletter, this is almost certainly not you yet. Buy Max only after Pro itself starts making you wait, because that’s the only proof you’d use what you’re paying for. And you’ll know. It’s very obvious when Pro isn’t enough.

The Two Clicks That Save You Money

Two more actions on that same screen, and both put money back in your pocket.

Flip to annual. Find the monthly/yearly toggle and click yearly. Watch Pro drop from $20 to about $17 a month. That’s roughly $40 a year for one click. Do it after your first full month on Pro, once you’re sure you’ll keep it, not before.

One toggle. $20 a month becomes $17. About $40 back in your pocket a year.

Know where the exit is. On the same manage screen there’s a downgrade and cancel option. Go find it now so you know it’s there. Plans aren’t a marriage. Try Pro for a month, and if the wall never comes back and you never touched Opus or Cowork, click right here and drop to Free.

The exit is right here. Plans aren’t a marriage.

Still Can’t Decide? Make Claude Decide

You’ve seen every screen now. If you’re still on the fence, hand the decision to Claude. Paste this into any chat, Free plan included:

Help me pick a Claude plan. Ask me 4 or 5 quick questions about how

often I use you, whether I've been hitting message limits, and whether

I need the smartest model or the hand-off features. Then recommend

Free, Pro, or Max and tell me in one line why. Don't upsell me.

It interviews you, then gives a straight answer based on your real habits. It’s got no commission to earn, so it’s a weirdly honest place to ask.

Hand it the decision. It interviews you about your real habits, then makes the call.

So, Which One Is You?

Here’s the whole decision in three lines, in order. Stop at the first that fits:

• Wall rarely shows up, don’t need Opus or the hand-off features? Stay Free.

• Hitting the wall weekly, or you want Opus / Projects / Cowork / Code? Get Pro. This is most of you.

• Blowing through Pro and feeling it in your work? Get Max, 5x first. Only touch 20x if 5x also runs out.

You climb one rung at a time, and each rung waits for the last one to hurt before you step up. That’s the trick to never overpaying. Let the wall tell you when to move.

A Couple of Gotchas

Paying more doesn’t make you better at this. A bigger plan buys room and a couple of models. It doesn’t do the aiming and steering from issue 2 for you. A sharp prompt on Free beats a lazy one on Max every time.

Max is two plans, not one. If you ever go there, start at 5x for $100. The jump to 20x is a real jump in price, and it’s only worth it once 5x itself isn’t enough.

Try It (Right Now)

If you clicked along, you’re basically done. Quick recap so it sticks:

1. Plan screen open? Bottom-left → Settings → Plan. That’s your control room.

2. Meter checked? On free, that’s one honest question: how often does the wall show up? Once a month is fine. Every couple of days means you’ve outgrown it. (On Pro or Max, just glance at Settings → Usage.)

3. Make the call. Free, Pro, or Max. If the honest answer is “Free’s fine,” that’s a win, you just kept your money. If it’s Pro, you now know exactly why.

The goal was never to get you to pay. It was to make sure that whatever you land on, you can say out loud why. Do that and you’ve beaten most people, who either overpay out of FOMO or sit on Free wondering why everyone else gets more out of it.

Next Week

Issue 4 is where the free stuff starts paying off big. We give Claude a real memory with Projects, so it knows your role, your goals, and your voice before you type a single word. It’s the difference between briefing a stranger every morning and walking in on a teammate who already gets you. Works on every plan, and it makes every prompt after it shorter.

See you then.

P.S. Are you Free, Pro, or Max right now, and does it feel like the right call? Hit reply and tell me. I’m genuinely curious where most of you land, and if a bunch of you are overpaying I want to say so out loud.

Much love ✌️ MJ

Last week you sent your first real prompt and got something done. An email rewritten, a PDF cracked open, an afternoon planned. That was the point: prove it works.

This week we make it land every time.

The gap between people who think AI is “meh” and people who swear by it comes down to the ask. Same Claude, same screen. One person types six words and gets mush. The other knows three small steps and gets back something they’d have paid a freelancer for.

Those three steps are the whole issue, and they go in order:

1. Aim it so you land it on the first try, without writing an essay.

2. Steer it when the first answer is close but not quite.

3. Teach it once, so it remembers and you stop repeating yourself.

No jargon, no “prompt engineering.” You’ll learn to say what you actually mean, then how to stop saying it twice.

The whole issue in one image: aim it, steer it, teach it.

Last Week’s Homework

Quick callback. Your job last week was to grab something real off your plate, send one prompt, then send one follow-up to fix it.

If you did it, you already used step number two without me telling you: you didn’t start over, you nudged. Hold that thought. It’s the spine of this whole issue.

If you skipped it, no judgment. You’ll do it now, because everything below is hands-on.

The One Mental Shift

Before the steps, the mindset they all come from. Get this and the rest is obvious.

Stop searching. Start briefing.

Most of us learned computers through Google. Type the fewest words possible, hope the machine guesses the rest. Two decades of training your brain to be terse. “flights paris cheap.” “chicken recipe easy.”

Claude is the opposite: a smart assistant waiting for instructions. And you’d never brief a real assistant in three words. You’d tell them what you need, who it’s for, and what “good” looks like. That’s all a great prompt is: a good brief.

So the upgrade is just saying more of what’s already in your head. You already know who the email is for and how you want it to feel. You’re just used to not bothering to type it. Start bothering. That’s most of the battle.

Step 1: Land It on the First Try (Without Writing an Essay)

Here’s the trap people fall into once they hear “add more context”: they write a 500-word prompt with their whole life story, and the answer gets worse, because the one detail that mattered is buried on line 14.

The goal is the right four or five things, and nothing else. Front-load the handful of facts that remove the guesswork, then stop typing.

Here are the levers. Pull the two or three that matter for your task and skip the rest.

Context that counts: who, what, good

The single biggest one. Claude can’t see your life, so it guesses generic unless you tell it three things:

• Who is this for? Your boss, a client, a five-year-old, yourself at 6am.

• What’s the situation? The one or two facts Claude would need to not say something dumb.

• What does “good” look like? Short and punchy, warm, formal, funny, blunt.

Watch it work. Same task, different ask:

Lazy: write a birthday message for my coworker.

Better: short, funny birthday message for my coworker Dana. She loves trail running and bad puns, we've worked together three years. Warm, not corny.

The first gets you a fortune cookie. The second gets you something Dana screenshots. And notice the “better” one is one sentence. One sentence with the right details beats a paragraph without them.

Tell it the shape

Claude will hand you a wall of text when you wanted three bullets, just because you didn’t say. So say, up front, and skip the “now make it shorter” round trip:

Give me three versions: short, medium, detailed.
Keep it under 100 words.
Put it in a table.
Just the final answer, no preamble.

That last line is a quiet favorite. It kills the “Great question! Here are some thoughts...” warm-up and gives you the good stuff straight.

Show it an example

This one feels like cheating because it works so well. Claude is a freakishly good mimic. One sample teaches it more than a paragraph of description.

Here's an email whose tone I loved: [paste it]. Write a new one telling a client we're two days late, same voice.

This is how you get it to sound like you instead of sounding like AI. Describing your voice in words is nearly impossible, so skip it. Show it and say “like this.” (That idea comes back big in step three.)

Hand it a role

Four words that reset the whole tone: “You’re a sharp copywriter.” “Act as a patient tutor.” “You’re my brutally honest editor.” A role is context for the voice, fast.

One prompt to steal

Here’s the whole step in a reusable shape. Five lines, not five hundred. Fill in what matters, delete what doesn’t:

I need you to [task]. It's for [who], and the situation is [one or two facts].

Make it [tone]. Format it as [bullets / email / table / one paragraph].

Here's what I'm working with: [paste your notes or an example].

That’s it. If you only ever learn this, you’re already past most people who’ve been clicking around for a year.

Swipe these (real tasks, already briefed)

Don’t just take my word for it. Here’s the template filled in for stuff you actually do. Copy one, swap the brackets for your details, send. Notice how each is a few lines, not a wall of text.

The weekly update you keep putting off:

Write my weekly status update for my manager. It's for a quick read,so keep it skimmable. Format: three bullets under "Done," three under "In progress," one line under "Need from you." Tone: confident, not boastful. Here's what happened this week: [dump your messy notes].

The message you’re dreading:

Help me tell a client we're running two days late. It's for [client], we've had a good relationship, and the delay is on us. Make it honest and calm, not a pile of excuses. One short paragraph I can send as-is.

Turning a meeting down without the awkwardness:

Write a polite decline to a meeting invite. It's for a colleague I like but the meeting isn't a good use of my time. Warm, brief, offer async instead. Two sentences.

Notes into something useful:

Turn these messy notes into a clean summary with clear action items and who owns each one. Bullets, no fluff, flag anything that's unclear:[paste your notes].

Explaining something to the right audience:

Explain [topic] to [who: my CEO / a new hire / my mom]. Match their level, skip the jargon, use one real-world analogy. Under 150 words.

Five prompts, five real jobs off your plate. That’s step one working.

The fortune cookie. Ask for nothing, get nothing.

One sentence, the right details. Same Claude, completely different result.

Ask for the shape, get the shape.

Step 2: When It’s Off, Nudge. Don’t Start Over.

Now the habit that separates the pros, and the one you already used in your homework.

When the answer isn’t quite right, almost everyone does the wrong thing: they delete it, rewrite the whole prompt, and start a fresh chat. Don’t. You just threw away everything Claude already learned about what you want.

Instead, talk to it like a person who’s 80% there. Just say what to fix.

Bookmark this nudge cheat-sheet. Almost every fix you’ll ever need is one of these:

Length:    "Cut it in half."  "Make it one sentence."  "Give me the longer version." Tone:      "Warmer."  "More casual."  "Sound like a human, lose the corporate voice."

More:      "That one's perfect, do three more like it."  "Give me other angles."

Structure: "Keep the structure, swap the examples."  "Turn this into bullets."

Redirect:  "You misread me, I meant the client, not my boss."  "Closer, but skip the intro."

Stuck:     "What would make this better?"  "Give me three directions you could take this."

Each nudge keeps everything good and fixes the one thing that was off. You’re in a conversation now: the first answer is a draft you react to.

And here’s the honest part the title doesn’t tell you: you won’t always nail it on the first try, and you don’t need to. Step one gets you to 80% fast. Step two closes the last 20% in one or two lines. That loop, brief then steer, is how every heavy user actually works. If you send one prompt and walk away disappointed, you quit one nudge too early.

The nudge, not the restart. One line fixes what was off and keeps what worked.

Step 3: Teach It Once, So You Stop Repeating Yourself

Now the step almost nobody knows, and the one that compounds.

Pay attention to your own nudges for a day and you’ll notice something: you keep sending the same ones. “No exclamation points.” “Give me three options, not one.” “Shorter.” “Drop the corporate tone.” Those are your standing preferences, and right now you’re re-teaching them in every single chat like it’s day one.

You don’t have to. Claude can remember.

There are two layers to this, and you’ve already met the first:

• Inside one chat, it remembers everything. Every nudge in a conversation sticks for the rest of that conversation. This is exactly why step two works, and why you keep one chat to one job.

• Across chats, you can make a lesson permanent. Tell it once, “from now on, never use exclamation points in my writing,” and it carries that forward into new chats. The nudge stops being something you repeat and becomes something it just does. This is built in, it works on every plan including free, and an instruction like that applies starting with your very next conversation. If it’s not already on, you’ll find the switch in Settings → Capabilities.

That’s the shift. Every correction you make can be a one-time investment instead of a daily tax. Teach it your voice once by showing it a few things you’ve written. Teach it your format once. Each lesson makes the tool a little more yours, so tomorrow’s first try is already closer to right than today’s was.

Three memories worth setting today

Next time you’re in a chat, try teaching it one of these. Paste it in on its own, and from then on it stops being a nudge you repeat:

Remember that I prefer short, direct writing with no exclamation points. 

Remember that whenever I ask for options, I always want three.

Remember that I'm a [your role] at [your company], so skip the basics when explaining work topics.

Set those three and half your daily nudges just disappear. Want to see what it’s holding onto? Settings → Capabilities → View and edit memory shows the full list in plain language, and you can edit or delete anything that’s gone stale. You’re never stuck with something it misremembered.

We’re keeping this at the concept level today, because doing it properly, with Projects, custom instructions, and memory that knows your whole world before you type a word, is the entire point of issue 4. For now, just know the door exists: when you catch yourself making the same nudge twice, that’s a lesson worth saving.

Teach it once. It sticks, so you stop repeating yourself.

A Couple of Gotchas

Small stuff, said out loud so it doesn’t bite you.

Relevant beats long. If a detail wouldn’t change the answer, leave it out. The skill is picking the details that matter.

One chat, one job. From last week, but it matters double here. Nudging a draft into shape? Keep it in that chat so the context stays clean. Start a new chat only for a genuinely different task.

It’ll ask if you let it. Not sure what to include? End with “ask me anything you need before you start.” Claude comes back with two or three questions, and answering them beats guessing.

Try It (Right Now)

Three minutes. One real task. You’re going to run all three steps on it.

1. Aim it. Take something you’d normally fire off in one lazy line. Rewrite it with the steal-this template: who it’s for, the situation, the tone, the shape. Send it. Notice how much closer it lands.

2. Steer it. However good that answer is, send one nudge. “Shorter.” “More casual.” “Two more like it.” Feel it sharpen.

3. Teach it. Whatever you just nudged, tell Claude to remember it. “From now on, keep my writing short and skip the exclamation points.” That’s the habit that pays you back all week.

Brief it, steer it, teach it. Do that and you’re the one driving.

Next Week

Issue 3: picking the right plan. Free, Pro, or Max, decoded without the upsell, so you buy exactly the Claude you’ll actually use and not a dollar more. If you’ve started bumping into the “you’re out of messages for now” wall, that issue has your name on it. And issue 4 is where memory goes from a neat trick to your default setup.

See you then.

P.S. What’s the same nudge you keep typing over and over? Reply and tell me. That repeated one is the exact thing you should be teaching it to remember, and I’m curious what everyone’s is.

Much love ✌️ MJ

You’ve heard about AI for two years now. The hype. The threads. The guy on LinkedIn (which is sometimes me) screaming that everything has changed.

This is the issue where you quietly skip all that and just use the thing.

It’s also issue 1 of a series. By the end of it, you’ll go from “I made an account” to running real work with AI: writing, research, building actual stuff, the kind of tasks that give you hours back. We’ll take it one step at a time, and every issue builds on the last. No prior experience needed. If you can send a text, you can do all of this.

Today’s job is small and specific. Give me 5 minutes. You’ll make an account, send a real prompt, and get back something you can use right now. A rewritten email. A reply you’ve been dreading. A plan for your afternoon. Then I’ll walk you through the corner of the screen that makes everybody freeze, so you never sit there wondering which button does what.

Coffee’s optional. Let’s go.

First, What Even Is This Thing?

Quick gut check before we click anything, because “AI assistant” means nothing until you’ve used one.

Claude is a chatbot made by a company called Anthropic. You type, it types back. So far, so boring. The part that matters is what it’s actually good at, which is a lot more than answering trivia.

Claude is genuinely great at:

• Writing and rewriting. Emails, posts, docs, that thing you’ve reworded nine times.

• Thinking through a problem with you. Decisions, plans, pros and cons, “talk me out of this.”

• Reading long stuff and telling you what matters. Paste in a contract, a report, a wall of notes.

• Explaining anything at the level you ask for. “Explain this like I’m five.” “Explain this like I have a finance degree.”

A few honest limits, so you’re not surprised later:

• It’s not a search engine by default. It knows a ton, but for fresh facts (today’s news, this week’s prices) you flip on web search. More on that below.

• It can be confidently wrong. It will occasionally state something false with a completely straight face. We’ll cover how to catch that.

• It starts each new chat fresh. It remembers everything inside one conversation, but open a brand-new chat and it doesn’t know what you said yesterday. We fix that in issue 4.

That’s the whole mental model. Smart, fast, a little overconfident, incredibly useful once you know how to talk to it. Let’s get you in.

Start in the Browser. Don’t Overthink It.

Skip the download for now. You don’t need it, and I’m not letting an install bar eat your 5 minutes.

Open a browser and go to claude.ai.

You’ll hit a sign-in screen. Take the fast lane:

• Continue with Google. Two clicks, no new password to forget by Thursday.

• Continue with email. Type your email, hit Continue, then grab the 6-digit code from your inbox.

If it asks for a phone number, give it one. Anthropic texts you a code. Pop in your first name, accept the terms, and you’re through the door.

That’s the whole signup. It’s free, and nobody’s asking for a credit card. You’ll start on the free plan, which is plenty for everything in this issue. We’ll talk about when it’s worth paying in issue 3, so don’t reach for your wallet yet.

Send Your First Prompt (and Please, Not “Hello”)

Almost everyone signs up and types “hi” like it’s a job interview. The poor thing says hi back, and you both stand there. Skip that. Hand it real work on the first swing.

Find an email you’ve been avoiding. Click into the big box at the bottom and paste this:

Rewrite this email to sound friendly but professional.

Give me three versions: short, medium, and detailed.

[paste your email here]

Hit Enter, or click the up-arrow (↑).

Now watch it write all three in real time. Version 2 a little stiff? Don’t start over like it’s a typo on a resume. Just tell it:

Make version 2 warmer and cut it by half.

That little back-and-forth is the whole game. You’re not searching. You’re nudging it until it’s right. We go deep on the nudging in issue 2.

Claude’s reply with the three email versions. Your first finished task. 👇

Don’t Like the Email Idea? Steal One of These.

The email rewrite is just the easy on-ramp. The point is to ask for something you actually needed done today. Pick whichever one sounds like your life and paste it in:

I have chicken, rice, and whatever's wilting in my fridge.

Give me three dinner options I can make in 30 minutes.

Plan my afternoon. I have these four things to do: [list them].

Tell me the smart order and roughly how long each should take.

Turn these messy notes into a clean summary with action items:

[paste your notes]

I need to tell a client we're running two days late.

Write it so it's honest, calm, and doesn't sound like an excuse.

Explain [thing you keep nodding along to but don't get]

like I'm smart but have never heard of it.

None of these are tricks. They’re just real. That’s the secret nobody puts on a slide: the people getting the most out of AI aren’t using magic words. They’re just asking it to do actual work.

One Thing That Separates a Good Answer From a Great One

Here’s a free preview of issue 2, because it’s too useful to make you wait.

Watch the difference between these two:

Lazy: write a birthday message for my coworker.

Better: write a short, funny birthday message for my coworker Dana.

She's on my team, loves trail running and bad puns, and we've worked

together for three years. Keep it warm, not corny.

Same task. Wildly different result. The first one gets you a fortune cookie. The second gets you something Dana actually screenshots. All you did was hand over context: who, what, the vibe you want.

That’s the entire skill in one sentence: tell it who it’s for, what you want, and how it should feel. Do that and you’ll already be ahead of most people who’ve been using this for a year.

The Bar at the Bottom, Decoded

Here’s the spot where new folks quietly panic. A row of little icons, zero explanation, and a creeping sense that clicking the wrong one will summon something. So let’s demystify the whole thing. Good news up top: today you touch exactly two.

The box in the middle is where you type. The up-arrow sends. Those two. That’s your whole job today.

Now the rest, left to right:

• The “+” button (bottom-left). The door to everything. Attach files, drop in photos, turn on Research, connect your apps. Up to 20 files per chat, 30 MB each. Hoarders, rejoice.

• The “/” shortcut. Type a slash for quick commands. A toy for later. Leave it alone today.

• The feather icon (Styles). Changes Claude’s writing tone. Also later. We’ll build you a custom one down the road.

  

Three of those “+” options change how hard Claude actually works on an answer. File these away. They’re your power tools once you’re rolling:

• Web Search. Flip it on when you need current facts. “What’s the weather in Cannes this week.” Live news. Anything that happened after breakfast.

• Extended Thinking. For the gnarly stuff that doesn’t need the internet. A messy decision. A bug. Anything where you want it to sit and actually think instead of blurting.

• Research. Hit “+” then Research and it goes off, reads a pile of sources, and brings back a real briefing in 1 to 3 minutes. The little Research icon lights up blue next to the “+” so you know it’s heads-down working.

One line to remember it all: thinking is for deliberation, search is for facts, research is for reports.

Drop In a File and Watch It Read

This is the moment most people realize Claude is different. You can hand it a document and it just reads the whole thing.

Try it. Click the “+” button, choose to attach a file, and grab something real off your computer. A PDF lease. A bank statement. A 40-page report you were supposed to read last week. Then ask:

Read this and give me the five things I actually need to know.

What should I be worried about?

A few seconds later you’ve got the gist of a document you’d been avoiding for a month. No skimming, no Ctrl+F, no pretending you read it.

This works on more than PDFs. Photos too. Snap a picture of a handwritten whiteboard and ask it to type up the notes. Screenshot a confusing error message and ask what it means. Photograph the inside of your fridge and ask what to cook. It can see, not just read.

Two ground rules. You can attach up to 20 files in one chat, 30 MB each. And don’t upload anything you’d be uncomfortable handing to a company, the same gut check you’d use before pasting something into any online tool.

Hand it the whole document.

“Wait, Which Claude Am I Even Talking To?”

Click the model name up top and a few options appear. This is where a lot of people short-circuit. Relax. There’s no wrong answer here.

Think of them as the same Claude on a dial. Smart and patient on one end, quick and breezy on the other.

Claude Sonnet is your daily driver. The best mix of smart and fast, and the one to reach for on almost everything. Drafting an email, summarizing a doc, planning your week. Honestly, you could stop reading this section.

Claude Opus is the big brain. Slower, but it’s who you call for the heavy lifting. Writing a real proposal from scratch. Working through a tangled decision. Anything where you’d happily wait an extra beat for a noticeably better answer.

Claude Haiku is the sprinter. Quick, simple, in and out. Great for fast little tasks where you don’t need it to ponder. “Fix the grammar in this sentence.” “Give me five subject line options.”

Your move for week one: pick Sonnet and leave it there. Only tap Opus when an answer feels a little thin and you want it to think harder. Done. We’ll match models to your plan in issue 3.

Sonnet as your daily driver. Opus for extra horsepower

Feeling like that’s a lot of buttons? It is. And you get to ignore almost all of them today. Type, send, leave it on default. The rest is just sitting there for when you’re ready, and we’ll unlock one piece per issue. No cramming.

A Few Things Nobody Tells You on Day One

Little stuff that trips people up in week one. None of it is hard once someone just says it out loud.

Your chats are saved. Look at the left side of the screen. Every conversation you start lives there in a list, so you can scroll back to that email you wrote on Tuesday. You can rename them too, which sounds boring until you have forty of them.

Start a new chat for a new topic. See the “New chat” button (usually top-left). Use it when you switch subjects. If you keep dinner ideas, work email, and tax questions all in one endless thread, the answers get muddier. One chat, one job. Cleaner every time.

It forgets between chats, for now. Inside one conversation, Claude remembers everything you’ve said. Open a fresh chat tomorrow and it’s a blank slate again. That feels annoying at first. It’s also fixable, and giving Claude a real memory is the entire point of issue 4.

It can be confidently wrong. This is the big one. Claude will sometimes state a “fact,” a quote, or a number that’s just made up, and it’ll sound completely sure of itself. So for anything that matters (a date, a stat, a legal detail, a name) verify it, or turn on web search so it’s working from real sources. Treat it like a brilliant, fast, slightly overconfident assistant. Trust it, then spot-check the things that count.

Free has limits. On the free plan you can send a good number of messages, then it asks you to wait a few hours before sending more. Totally normal. If you start hitting that wall a lot, that’s the signal you’re ready for issue 3, where we sort out plans.

Now Make It Permanent

You’ve felt it work. Time to put Claude somewhere you’ll actually reach for it.

On your computer. Go to claude.com/download and click macOS or Windows. Open the file, install, launch it from your Applications folder (Mac) or Start menu (Windows). Sign in with the same account and your chats are already there waiting. Everything syncs, so the email you wrote on your laptop is on your phone by the time you’ve stood up. (Get the desktop app. Cowork lives here, and that’s a whole issue I can’t wait to write.)

On your phone. Open the App Store or Google Play, search “Claude by Anthropic,” install, sign in. Done. Little bonus: tap the sound-wave icon next to the mic and you can just talk to it out loud. Voice mode is free on mobile, and it’s weirdly fun on a walk. I work through half my ideas talking to it like a coworker who never gets tired of me.

📸 [SCREENSHOT 9 — claude.com/download with the macOS and Windows buttons. Annotation: circle both.] 📸 [SCREENSHOT 10 — The Claude app in the App Store, showing the “Claude by Anthropic” publisher. Annotation: “Make sure it’s the official one.”]

Your Homework

One task. Right now. Before you close the tab and “get to it later,” which we both know means never.

Grab something real off your plate. The reply you keep flinching at. A paragraph that won’t behave. The PDF you’ve been ignoring. “Plan my afternoon around these four things.” Hand it over and send one prompt. If the first answer isn’t quite right, send one follow-up and fix it. That second message is where it clicks for most people.

That’s the whole assignment. Congrats, you’ve officially used AI for real work instead of asking it to write a poem about your dog.

Next week, issue 2, we make every prompt land. The few small habits that turn a decent answer into the exact thing you had in your head. You got a taste of it with the Dana example. We’re going to make that automatic.

See you then.

P.S. What was the first thing you actually asked it? Hit reply and tell me. I read every single one, and the weird ones are my favorite.

Much love ✌️ MJ

A list of where they worked. A headshot from 2019. An About section copied off their company bio. They set it up once and never look at it again.

That’s a mistake, and it’s costing them.

Your profile is the most-visited page you own. Every time you post and someone wants to know who you are, they land here. Every time your name shows up in a comment, a search, a DM thread, people click. They’re deciding in about three seconds whether to follow, subscribe, or scroll on.

I have 70,876 followers. When a post lands, thousands of people click my name and end up on my profile. That page is what turns attention into pipeline. Not in theory. In practice.

It’s a landing page. And like any landing page, every section has a job.

Today I’m going to walk you through mine, section by section, with the exact specs. Five sections. Steal what works.

Section 1: The Profile

This is the fold. Banner, photo, headline, button. Everything here decides whether someone stays long enough to read the rest.

Banner (1584 x 396px). Treat it like a billboard, not a backdrop. One clear headline. Mine reads “Build your brand & business on LinkedIn.” Under it, social proof: logos people recognize so they trust me before they’ve read a word. And a large photo of myself, because people follow people, not gradients. Keep it concise. A banner crammed with six bullet points reads like a slide nobody asked for.

Profile photo (1000 x 1000px). Simple background. Clear. Professional. Smile. That’s the whole list. Your face is the most-clicked pixel on the page, so don’t waste it on a logo or a group shot where I have to guess which one is you.

Headline. This is the line the algorithm reads first, and the line that follows you across every comment and search you ever show up in. There’s a formula:

[who you are] + [action verb] + [ideal customer] + [what they want] + [how you help]

Mine: “Build an operator brand — AI content that sells. | GM @ Stanley.” The credential sits at the end on purpose. It earns the right to the claim. Write yours so a stranger reads it once and knows exactly who you help and what they get.

Custom button. Most people leave this on the default “Connect” or “Message” and never think about it. Change it. Point it at your primary offer. Mine says “View my newsletter.” One click from my name to the thing I actually want people to do.

Subscribe now

Section 2: Featured

Below the fold, this is your highest-value offer. Not your highest-performing post. Big difference.

Most people pin their viral post here. A viral post is a dead end. It got attention, and then what? Nowhere to go. The Featured section should point to something you’re building or selling, something that moves a stranger from “I like this guy’s posts” to “I’m on his list” or “I bought his thing.”

I run two: my AI-Proof newsletter and my LinkedIn Design Handbook. One captures the email, one is a product. Both have custom thumbnails, because the default link preview looks like exactly what it is, a default link preview. Design the cover. It’s the difference between a billboard and a sticky note.

Section 3: About

This is where most operators sound like a press release written by committee. “Results-driven leader with a passion for leveraging synergies.” Nobody talks like that. Nobody reads past it either.

Write like a human. Start conversational. Mine opens with “Hey, I’m MJ.” Then name the problem your reader actually has. Mine: most operators are invisible to the market they sell into. From there, establish credibility fast with a specific result, not a buzzword. Then bridge to what you do now, and end with a clean invitation to connect.

The order matters: open warm, name their problem, prove you’ve done the thing, tell them what you do now, invite them in. Show the receipt with a real number somewhere in there. Cut every adjective that isn’t carrying weight.

Section 4: Activity

This is your real resume. Not the Experience section underneath it. This one.

Activity is the proof you actually do the thing you claim to do. If your headline says you help founders grow with content and your last post was eight months ago, the profile lies, and people can tell. When someone lands on your profile and sees a wall of recent, useful posts, you’ve already won them. They don’t need to read your job history. The consistency is the credential.

So post consistently. Show what you’ve built. Show who you help. Your Activity feed is the only section that proves the rest of the page is true.

Section 5: Experience

Keep it punchy. Show impact with real numbers.

“GM at Stanley” is fine. “GM at Stanley, leading an AI product for LinkedIn creators” is better. A role with one line about what you’ve actually driven beats a six-bullet job description every time. Nobody scans LinkedIn looking to read your responsibilities. They’re hunting for the one number that tells them you’re legit. Give them that number per role and cut the rest.

The Page You Own

Every post you write is a door. The profile is the room people walk into when they open one.

You can spend all your energy on the doors, and plenty of people do. But if the room is a mess, none of it converts. Build the room first.

Open your profile right now. Start at the banner and work down. Pick one section and fix it this weekend. Then the next. You’ll see the difference in who follows, who subscribes, and who reaches out.

Thanks for reading. If this was useful, share it with an operator who’s still treating their profile like a resume.

Much love ✌️

MJ

I want to talk about a category that’s been quietly winning for two decades, and that almost nobody calls by its real name.

Operator brands.

People who run something, whether a company or a team or a function with real revenue attached. They publish to compound that thing. The audience is the leverage on the business they’re already running. They aren’t trying to become creators. They’ve never thought of themselves that way. They publish because publishing is part of how the business gets built now.

You don’t have to become a creator to compound your business. You just have to publish like one.

That’s the line I keep coming back to with operators I meet, and most of them still flinch when they hear it. They think publishing is for influencers. They’re building real things. Software, agencies, consumer brands, funds. They don’t see what their inner LinkedIn poster has to do with any of it.

I get the resistance. I had it myself for a long time. But while you sit on it, somebody less qualified than you is publishing about your category every week, and turning that publishing into pipeline that should have been yours. That cost compounds quietly. Most operators don’t notice it until they’re trying to launch the next thing and realize they’re starting from zero.

I run this play myself. I’m GM of Stanley. I also own and operate a small retail store in Manhattan where I live. I’ve been an operator, executive, founder, and business owner for most of my career. I started publishing because I was obsessed. Not because I had a plan. The content isn’t a separate career for me. It’s the receipts of the businesses I’m running, told out loud, so the next person who needs what we sell already knows who I am before we meet.

This pattern isn’t new and it isn’t an AI cycle thing. Operators who publish have been winning their categories for twenty years. They were just rarer. Once you see the pattern once, you start seeing it everywhere.

The 20-year precedent

Gary V (2006)

Gary Vaynerchuk did not show up on YouTube with nothing to show.

By the time he picked up a camera, he’d already taken his family’s wine business in New Jersey from $3 million to $45 million. He started at fourteen, bagging ice for $2 an hour. Took over operations. Rebranded the shop Wine Library and ran one of the first wine e-commerce sites in the country.

Then on February 21, 2006, he hit record on Episode 1 of WineLibraryTV. Daily video about wine, on early YouTube, at a time when most businesses thought video on the internet was a stunt. He kept producing through 2011. At peak the show drew about 100,000 views per episode. By 2009 he’d co-founded VaynerMedia with his brother AJ, and by 2013 the agency was doing $23 million in revenue.

The order is the whole point. Wine business made him an operator. WineLibraryTV gave him a public identity. The identity made the agency possible. He kept publishing through every chapter of all of it.

People remember the loud part: the keynotes, the hashtags, the Gary Vee branding. Worth remembering the quiet part too. He spent a decade building distribution that paid him back every time he wanted to start a new thing.

Andreessen Horowitz (2009)

Same year Gary started VaynerMedia, two operators started a venture firm.

Marc Andreessen had co-founded Netscape. Ben Horowitz had run Opsware as CEO and sold it to HP for $1.6 billion in 2007. Both were already publishing before they ever wrote a check. Marc’s “Pmarca Guide to Startups” went up in 2007, two full years before a16z existed, and it became the closest thing the early Web 2.0 generation had to a founder’s bible. Why not to do a startup. The only thing that matters. How to hire executives. Founders quoted those essays then. People still quote them now.

When the firm launched in July 2009, content was load-bearing from day one. They hired Margit Wennmachers to run media. Ben started bhorowitz.com around 2010 and wrote raw operator-CEO posts in a voice no other VC was using at the time, opening posts with rap lyrics, telling the truth about firing executives, talking about the struggle. The earliest archived post, Why We Prefer Founding CEOs (April 28, 2010), reads like a memo from somebody who had actually run a public company and had the scars to prove it.

That body of work eventually became The Hard Thing About Hard Things in 2014. By then Sonal Chokshi had come over from Wired to build out the podcast and the rest of the publishing arm. Hundreds of episodes followed. By the time the rest of the venture industry caught up and decided every fund needed a media operation, a16z had been running theirs for half a decade and was sitting on the brand position that helped them compound to roughly $28 billion in AUM.

The operators built the receipts first. They started publishing earlier than their peers. The brand became the unfair advantage that’s still paying out today.

The pattern again, AI-native edition

That was 2006 to 2009. The cycle’s running again right now, faster than before, because the team sizes are smaller and the audiences are easier to reach.

Lovable

Anton Osika has been treating Lovable’s growth as the story of Lovable’s growth. Public launch in November 2024, $4 million ARR in four weeks, $10 million ARR in sixty days with fifteen people on the team. By July 2025 the number had crossed $100 million, and by November it was $200 million. Every milestone is also a post. Anton writes the number, writes how they got there, writes what almost broke along the way. The product would be growing either way. What the publishing does is make Lovable the company everyone is already watching when it ships next, which is the part of the flywheel that competitors with quieter founders can’t reproduce no matter how good the product is.

Gamma

Grant Lee runs Gamma with around fifty people, serves roughly seventy million users, and has been profitable for more than a year straight. The interesting part isn’t the lean team. It’s that Grant publishes the growth playbook openly. He writes about how they launched features in a day, how they hired, what worked, what didn’t. Other operators read it and copy it. Some of those operators end up buying Gamma because by the time they need a presentation tool, they already think of Gamma as the company that thinks about decks correctly. Teach the playbook well enough and you become the playbook.

Clay

Kareem Amin and Varun Anand named the category they sell into. Two years ago they were calling a role “GTM engineer” inside their own Slack. Now people put it on their LinkedIn titles, the median salary is around $160K, and universities are starting to build curriculum around it. Clay sits at the center of that category because Clay drew the category lines, largely in public, while they were still small enough that nobody had to take them seriously. When you draw the lane, the lane runs through you.

And it doesn’t only show up in AI

The pattern is bigger than any one cycle. Adam Robinson runs RB2B and has turned radical revenue transparency into the moat that defines his corner of SaaS. Cody Plofker is building Jones Road into a serious DTC brand while publishing the playbook of how DTC actually works in 2026. Tobi Lütke has been doing this at Shopify for over a decade. Brian Chesky’s “founder mode” essay reframed how every Series B founder I talk to thinks about company-building. Aaron Levie was building Box’s operator brand back when the idea was still considered exotic in enterprise.

Different industries, different decades, same shape underneath.

Why operators get confused with influencers

The reason most operators stay quiet, in my experience, is they think publishing is the same job as being an influencer. It isn’t, and the difference matters a lot for how you should think about your time.

For an influencer, the content is the product. They wake up every day to grow the audience. Their scoreboard is followers, views, brand deals. Their risk is mostly the algorithm. If the platform changes, the business changes with it. Their receipts are usually that they read about the thing they’re teaching.

For an operator, the content is fuel for the product. The scoreboard stays revenue, pipeline, and repeat customers. The risk is somewhere else, because the business doesn’t depend on the algorithm to function. The receipts are that you ran the playbook. You can tell people what it actually cost and what it actually returned, because you were the one writing the checks.

That difference is why pure creator economics look the way they do. A small number of people at the top of the curve get rich on courses and brand deals. Most of the rest of the people calling themselves creators make less than a junior employee at the kind of company they’re trying to teach. Same hours into content, very different ceiling.

The operator who publishes is playing for a different ceiling, which is equity in a real business, recurring revenue, an eventual exit, and IP you keep. Same hours into content, very different math at the end.

Pick which game you’re playing on purpose, because the two paths diverge quickly and you can’t run both at full speed.

What makes the move actually work

I’ve watched enough operator brands compound, and enough fail to compound, to think there’s a short list of things that have to be true if any of this is going to work for you.

The first is that the receipts come first. People can smell content written by somebody who’s never run the thing they’re talking about, and the moment they smell it your credibility evaporates. Build the thing. Publish what’s true about building it.

The second is to move early in your category. Publish before publishing is normal in your lane. By the time it’s normal, the people who started early have already compounded their distribution and your timing is mediocre.

The third is to treat content as a moat instead of a marketing channel. Marketing is something you do this quarter. A moat is what you build for ten years and then sit on for the rest of your career. The math is completely different and the kind of content you make should be too.

The fourth is to write like a person. The reader is choosing to spend two minutes with you because they want a human’s take, not a press release. The moment the writing starts to feel produced, they’re gone.

The fifth is that the building should be the content. If you have to invent a content strategy that’s separate from what’s actually happening at your company, you’ve already lost the thread. The interesting stuff is what you’re already doing. Tell that honestly.

Where I’m running this

The day job for me right now is Stanley, an AI tool for LinkedIn content creation that I’m trying to get from a couple million in ARR to $100M. The retail shop is its own thing and I love running it. The LinkedIn account, this Substack, and the Instagram account I’m spinning up sit on top of all of it. The point is that whatever I build next starts the day it starts with an audience of operators who already trust me.

That’s the asset I’m building. The point isn’t fame. The point is that I’d like to never start from zero again.

If you operate and you’ve been quiet, this is a fine moment to stop being quiet. Publish before publishing is normal in your category. Use the receipts you already have. The lane is wide open and almost nobody in your category is in it.

You’re a builder. Build your brand the same way you build everything else.

If it’s to be, it’s up to you.

Save this. Send it to the operator in your life who keeps telling you LinkedIn is cringe.

P.S. What’s the real reason you haven’t started publishing yet?

Much love ✌️ MJ

I used to spend 30 minutes to an hour on a single infographic.

It was time-consuming. It was painstaking. And it was a bottleneck.

Multiply that by three graphics a week and I was burning hours just to publish content I’d already written.

So I stopped making graphics. I built a system that makes them for me.

Now I describe what I want in plain English, and Claude Cowork hands me back an on-brand graphic that matches my colors, my fonts, my border style, my footer. No Canva. No fiddling. No “close enough.”

This isn’t a prompt I copy-paste. It’s a setup. Once you build it, every future graphic gets faster, more consistent, and more yours.

Here’s the exact 6-step process I use to turn Claude Cowork into a personal design studio.

Step 1: Start a project and connect a folder

Most people open Claude, type a prompt, and close the tab. That’s a chat. We’re building something different.

A project in Claude Cowork is a persistent workspace. It remembers your files, your instructions, and your context every time you come back. If you don’t set this up, you’re starting from zero every session and your “system” is really just you copy-pasting the same brand guidelines into a fresh chat over and over.

Open Claude Cowork. Click New Project. Name it something specific like “LinkedIn Graphics” or “Substack Visuals.” Don’t call it “Design” or you’ll forget what it’s for in three weeks.

Then connect a folder from your computer. This is the part people skip and regret. The folder is where your brand assets, reference posts, and finished graphics live. It’s also where Claude saves the editable files it produces. Without a connected folder, every output evaporates when the session ends.

Create a dedicated folder for this. Something like ~/Documents/Claude/Brand Designs. Drop in subfolders for brand-kit, references, and outputs so things stay organized as you scale.

This is the foundation. Skip it and the rest of the system has nowhere to live.

Step 2: Feed it your real brand file

Most AI design tools fail here. They guess at your brand. Close-ish colors. A font that looks like your font. A vibe that’s almost right.

That’s why the output looks generic.

The fix is simple. Give Claude the source of truth. Upload your actual brand kit. If you use Canva, export your brand kit as a PDF and drop it in. If you have a brand guidelines doc, use that. If you have neither, take ten minutes and make one: your hex codes, your fonts, your logo, your border style, your footer treatment.

Then write this prompt:

Pull my exact colors, fonts, and border size from this file. Don’t guess. Save them to a file called brand-tokens.md in this project so you can reference them every time we make a graphic.

What you’ve just done is taught Claude your brand once, in a way it can reuse forever. Every future request can pull from brand-tokens.md instead of you re-explaining “the background should be cream, not off-white” for the hundredth time.

Test it. Ask Claude to list your brand colors back to you with their exact hex codes. If it gets them right, you’re locked in. If it hallucinates, point at the file again and tell it to read more carefully.

This step is the single biggest difference between AI design that looks like everyone else’s and AI design that looks like yours.

Step 3: Show it a post you love

Telling Claude your brand colors isn’t enough. It needs to see the finished product. The same way a new designer needs to see your existing work to match the style.

Pick one graphic you’ve published that you love. The one that got the most engagement. The one where every detail is right. Upload it to the project.

Subscribe now

Then point at the parts that matter:

The border, footer, and corner radius on this graphic have to match exactly in everything you make for me. The title typography sets the tone. Use this as the visual reference for the system.

You’re not asking Claude to copy this graphic. You’re calibrating its eye. From here on, when you ask for a new graphic, it has a reference for what “right” looks like instead of generating something generic and waiting for you to redirect it.

If you have a few graphics in different formats (carousel, single image, header), upload one of each. Label them so Claude knows what they’re for. “Reference for single-image LinkedIn post.” “Reference for carousel slide 1.”

The principle: AI is great at matching patterns. It’s terrible at inventing your taste from scratch. Show it what you’ve already decided looks good, and it stops guessing.

One reference graphic upfront saves you twenty rounds of “no, more like this” later.

Step 4: Ask for a system, not a graphic

Here’s where most people stop short. They ask for a graphic. They get a graphic. They love it. Then next week, they have to start over.

That’s not a system. That’s a one-off.

The shift is asking Claude to build a reusable template instead. Something where the brand styling lives in one file, the layout lives in another, and the words on the graphic live in a third. So when you want a new graphic next week, you only change the words.

Try this prompt:

Build me a reusable template for single-image LinkedIn graphics. Brand colors and fonts in one file. Layout structure in another. The actual headline and body copy as data I can swap. Output an editable HTML or SVG file plus a finished PNG so I can both ship and tweak.

What you get back is a tiny design system. Three files. Brand. Layout. Content. The brand file rarely changes. The layout file changes when you want a new format. The content file changes every time you publish.

This is the difference between getting a fish and learning to fish, except you’re the one teaching Claude to fish for you.

The first time you do this, it takes longer than just making one graphic. The tenth time, you’re producing a new on-brand graphic in two minutes flat. That’s the trade.

Build the machine once. Use it forever.

Step 5: Edit one sentence at a time

The first version is almost never right. That’s fine. The system is built for iteration, not perfection on the first try.

The mistake people make here is giving Claude a paragraph of feedback at once. “Make the title bigger, square the corners, the background is too dark, the footer should be smaller, and the icon needs to be on the right not the left.” Claude tries to do all of it and botches half.

Instead, give it one note at a time. Plain language. Specific.

Square the corners.

Check the output. Then:

Make the texture more subtle.

Check again. Then:

Bigger title. Tighter line height.

Each note, one change. Each output, a quick check. You’re not writing a design brief. You’re nudging a system that’s already 90% there.

This is also how you train the system over time. When you keep saying “square the corners,” eventually you tell Claude to save that as a default in the brand file. Next week, no more square-the-corners note. The system absorbs your preferences and stops needing them.

This is the part that feels slow at first and ridiculously fast within a month. Every preference you teach the system is one you never have to teach it again. Your future self is doing less work because your current self is being specific.

The system gets sharper every time you use it.

Step 6: What it makes (the output)

Here’s the part nobody believes until they see it.

After 30 minutes of setup, you can ask for a new on-brand graphic with a one-sentence prompt and get a finished image back in under two minutes.

New single-image graphic. Headline: “10x Your Claude Outputs.” Six numbered tips. Use the template.

That’s it. No Canva. No font selection. No color picking. No alignment. Claude pulls your brand tokens from Step 2, your reference style from Step 3, your reusable template from Step 4, and applies the preferences you’ve trained into it through Step 5.

What lands in your output folder is two files. A PNG ready to post. And the editable source file in case you want to tweak something.

The infographic you’re reading this newsletter about? That’s a real output from this exact system. I described it in one paragraph. Claude built it. I made two small edits. Done.

Now multiply that by every graphic you make for the rest of the year.

This is why you build the system instead of the graphic. One graphic saves you nothing. A system that makes every graphic for the rest of your life is a different kind of leverage.

The shift

Most people use AI to make things faster. That’s the small version of the play.

The bigger version is using AI to build systems that make things for you. The graphic is the deliverable. The system is the asset.

Set this up once and you stop being the person who makes graphics. You become the person who designs the machine that makes graphics. Your time goes to strategy, writing, talking to your audience. The visual production layer runs itself.

Build it this weekend. Start with one format. Get it dialed in. Then add the next one.

If you set this up, send me a DM on LinkedIn with the first graphic your system produces. I want to see it.

Much love ✌️

MJ

I used to walk into Mondays blind.

9am Slack from my boss about something we’d talked through on Friday. I’d nod along in my head, scroll up to find the thread, fail, then stumble my way through a response. By 10am I had three meetings stacked and zero memory of where we’d left off in any of them. Around 2pm I’d remember the LinkedIn post I never queued for Tuesday.

The cost wasn’t time. It was making decisions on stale context.

This is the part nobody talks about with AI assistants. Most people treat them like a chatbot you go visit. Type a question, get an answer, close the tab. That’s the wrong frame. The actual unlock is scheduled work that runs before you wake up.

I set up a Daily Brief in Claude Cowork a few weeks back. It runs every morning at 7am. By the time I’m pouring my first coffee, the briefing is already sitting there. Meetings, with context. Overdue items I forgot existed. Gaps in next week’s content I would’ve missed.

It took me 10 minutes to build. It’s been giving me back the first 45 minutes of my day ever since.

Not in theory. In practice.

Here’s exactly how to do it.

How to Set Up Your Daily Brief in 10 Minutes

No fluff. This is the whole thing.

Step 1: Download Claude Cowork (2 minutes)

Go to claude.com/download and grab the desktop app for your OS. Sign in. Connect a working folder. Pick whichever one holds the files you actually touch every day.

You need the desktop app. Cowork doesn’t run in the browser, and the scheduled-skills piece (the whole point of this) only works in the native app.

Step 2: Connect Four Tools

The brief is only as good as the data it pulls from. You need four connectors, no more for now:

Granola. For meeting notes. So Cowork can pull what was actually said the last time you talked to someone.

Notion. For your content calendar and to-do list. So it knows what’s overdue and where the gaps are.

Gmail. For conversation context. So it can summarize the latest thread with whoever you’re meeting today.

Google Calendar. For the day ahead. So it knows who you’re seeing and when.

Resist the urge to wire up everything on day one. Get these four working. Add more later.

Subscribe now

Step 3: Build the Skill (3 minutes)

In Claude, go to Skills, then Create New. Paste this exact prompt:

Every morning at 7am, give me a daily brief with three sections:

(1) Content Calendar Review: flag gaps in the next 7 days.

(2) Overdue Tasks: surface items past due.

(3) Meeting Prep: for each meeting today, pull Granola notes from the last conversation and summarize the most recent email thread. One talking point + one open thread per meeting.

That’s it. No prompt engineering wizardry. Specific instructions, three sections, clear scope.

Step 4: Schedule It for 7am

Set the skill to run automatically at 7am every day.

This is the unlock. The whole reason it works is that it runs before you start your day, not when you remember to ask for it. If you have to remember to trigger it, you won’t. Schedule it once and forget about it.

10 minutes. Done. The brief now writes itself every morning, forever.

Why This Beats Reading More

Most people don’t lose to better operators. They lose to better-prepared ones.

I used to think preparation meant reading more. More inbox time. More scrolling through Notion. More open tabs. That doesn’t work because the surface area keeps growing. Inbox is bottomless. Notion always has more pages. Calendar invites keep coming.

The Daily Brief flips it. Instead of you hunting through five tools every morning, one tool hunts for you. It surfaces what matters. You act on it. You close the loop. Then you get to the actual work.

Set it up this weekend. It’ll take you about as long as one episode of whatever you have on in the background.

Tomorrow morning at 7am, you’ll get the first one.

You’ll wonder how you ran your week without it.

Happy Sunday

MJ ✌️

Everyone needs to be using Claude.

If you’re not, you’re already behind.

That doesn’t mean lighting your money on fire to use it.

Most people hit Claude’s usage limit and blame the limit.

The limit is fine. The way you’re using Claude is the problem.

I hit the ceiling once. Mid-task, mid-week, mid-flow. Locked out for hours.

So I sat down and figured out where my credits were actually going. Not what I assumed. What was real.

Turns out 80% of my usage was coming from habits I didn’t even notice. Bloated uploads. Stale chats I never reset. Features running in the background I wasn’t using.

I rebuilt the way I work with Claude around 9 specific moves. I haven’t hit the limit since. Not in theory. In practice.

Here’s the playbook.

9 Ways to Never Hit Claude’s Usage Limits

No fluff. Just what actually works.

1. Right Model, Right Job

Not every task needs Opus. Most don’t.

Cleanup work, formatting, summaries, simple rewrites. Haiku handles those at a fraction of the cost.

Drafting, analysis, planning. Sonnet is the workhorse.

Hard reasoning, complex code, multi-step strategy. That’s when you reach for Opus.

People default to the most expensive model and wonder why they run out. Picking the right tool for the job is free leverage.

Summary: Right tool. Right task. Right cost.

2. Kill the Extras

Web search. Connectors. Extended thinking.

Every feature you leave on burns credits even when you don’t use it.

Writing a LinkedIn post? No reason for web search to be on.

Drafting an email reply? You don’t need extended thinking.

Working from a doc you already pasted in? Connectors can be off.

Treat each chat like a clean kitchen. Pull out the tools you need. Leave the rest in the drawer.

Summary: Pay for what you touch. Nothing else.

3. Trim the Fat on Uploads

PDFs and screenshots are credit vampires.

That 40-page PDF you uploaded? Claude is processing every word, every footer, every page number. Even if you only care about page 12.

Same with screenshots. You only needed the text in the middle. Claude is parsing the whole image.

Paste the relevant text into a plain message. Crop screenshots to the part that actually matters.

You’ll cut your token cost in half on most file-based tasks.

Summary: Lighter files. Lighter bill.

4. Park Recurring Files in Projects

If you’re re-uploading the same deck, brand guide, or reference doc into every chat, you’re paying for it every time.

Drop it into a Project once. Every chat inside that Project has it loaded automatically. No re-upload. No re-processing. No re-paying.

I keep my voice guide, content pillars, and audience profile in a “Content” Project. Every post I write starts with that context for free.

Summary: Load once. Reuse forever.

5. Scope Before You Build

Don’t open a fresh chat with “make me a spreadsheet that tracks my LinkedIn posts.”

That’s how you end up with 5 versions, 4 rewrites, and a credit bill that should’ve been one clean build.

Use Chat to scope first. What columns? What sources? What format?

Lock it down in conversation. Then ask Claude to build it once the spec is tight.

Cheap conversation upfront. Expensive output once. Not the other way around.

Summary: Cheap thinking. Expensive output.

6. Make Claude Interview You

Stop writing 500-word prompts.

End your message with: “Ask me questions first.”

Then click the choices Claude offers.

You’ll get a sharper output in a fraction of the words. Clicks are free. Typing isn’t. And every word you typed was a token Claude had to process before it even started working.

This is the single fastest fix on the list. Try it on your next big task.

Summary: Clicks are free. Typing isn’t.

7. Edit. Don’t Repeat Yourself.

Claude flubbed the output. What do most people do?

Send a follow-up: “Actually, can you redo it but this time...”

Now Claude reads the original prompt, the bad output, your correction, and generates a new version. You just paid for the same task twice.

The fix: edit the original message. Regenerate. The bad version disappears. The new one replaces it.

One charge, not three.

Summary: Every “actually, I meant...” jacks up the tab.

8. Reset the Chat Every 10-15 Messages

Long threads drain credits.

Every message you send re-reads the entire chat history. By message 30, Claude is re-processing everything you’ve ever said in that conversation just to answer “what about this?”

Hit a natural breakpoint? Summarize where you are. Copy what you need. Start fresh.

Two short chats almost always cost less than one long one.

Summary: Short chats = cheap chats.

9. Space Out Your Sessions

Claude runs on a rolling 5-hour window. Burn it all in hour one and you’re locked out for four.

Plan your work in 2-3 sessions across the day instead of one big push. Combine asks inside each session so you’re not opening a new chat for every micro-task.

By the time you’d be hitting the limit on a single sprint, your usage from session one has already reset.

Summary: By the time you’re back, your limit already reset.

The Real Move

Hitting Claude’s limit isn’t a Claude problem. It’s a workflow problem.

Most people use Claude like a slot machine. Pull, pull, pull, hope something good comes out.

The operators I know use it like a tool. Right model. Right scope. Right files. Right pacing. The output is better and the bill is half.

Pick three of these to install this week. Watch what happens to your usage.

Thanks for reading. If this saved you a session, share it with someone who keeps hitting the wall.

Much love ✌️

MJ

Two months ago, I used Claude as a fancy search engine. Prompt, hope, repeat.

That is 10% of what it can do.

Things got interesting when I stopped treating Claude as a chatbot and started treating it as a system that gets work done for me. I trained it on who I am, how I think, my goals, and how I write. The most important part: I trained it on how I work, how I make decisions, and what I need done each day.

The outputs are way better. I prompt less. Work actually gets done.

It runs on four files. Twenty minutes to set them up. Roughly two hours a day back, every day.

Not in theory. In practice. Here is the system.

1. Agents.md

Your AI org chart.

Agents.md defines who works for you. Each agent has a role, a personality, and clear boundaries. The Email Agent does not write content. The Content Agent does not triage your inbox. They stay in their lane.

That sounds obvious. It is not how most people use Claude.

Most people throw every task at one generalist and wonder why the output feels mid. Specialists beat generalists, every time. The same way you would not ask your accountant to design a logo, you should not ask one undefined agent to do everything.

My setup right now:

• Email Agent. Triages my inbox and drafts replies in my voice.

• Brief Agent. Preps me for every meeting before I walk in.

• Content Agent. Fills calendar gaps with posts ready to ship.

• Ops Agent. Keeps Stanley and Rock Star Crystals running between weekly check-ins.

Define the agent once. It knows its job forever.

Subscribe now

2. Context.md

Your company brain.

Context.md is the file Claude reads first, every single session. It holds your business overview, your voice rules, your key decisions, the tools you use, and the guardrails you operate inside. Anything Claude needs to feel experienced instead of starting from scratch.

This is what stops you from re-explaining your business every Monday. New chat, new project, new agent, and Context loads first so Claude already knows where it is.

What goes in mine:

• A one-paragraph business overview (what I do, who I serve, how I make money)

• Voice and tone rules (do this, never that)

• Active products, projects, and priorities for the quarter

• The tools I use and how they connect

• Decisions I have already made, so Claude stops re-asking

Without Context, Claude is a smart stranger. With it, Claude is a teammate who already knows the playbook.

3. Memory.md

Where Claude learns you.

Memory.md is what makes session 100 smarter than session 1. Every correction, every preference, every win, every mistake gets logged. The next time you sit down, Claude already knows.

You said your name is pronounced a certain way. Claude logs it. You corrected the tone of a draft. Claude logs it. You picked option B over option A in a decision. Claude logs the why.

No re-explaining. No starting over. Compounding knowledge.

This is the file most people skip and the one that pays off the longest. Every other file is static. Memory is alive. It is the difference between a tool you reset every Monday and a teammate who has been with you a year.

4. Skills.md

Your reusable playbooks.

Skills.md is where your best processes live. Draft a LinkedIn post, triage Slack, pull the weekly numbers, prep a board update. Write the steps once. Any agent can run them.

This is the difference between AI that improvises and AI that operates. Improvising is fun for a demo. Operating is what saves you ten hours a week.

The rule: if you do it more than three times, it is a Skill.

What lives in mine:

• /linkedin (rough idea in, post-ready draft out)

• /weekly-report (sales, content, and ops in one document)

• /meeting-brief (calendar event in, full prep doc out)

• /inbox-zero (Gmail in, sorted action list out)

SOPs die in Google Docs. Nobody opens them. Skills run on autopilot. Slash command, done.

Build once. Run forever.

How They Work Together

Context loads first. The foundation.

Agents read Context and know their role.

Agents run Skills to execute real work.

Memory captures what happened.

Next session starts smarter than the last.

Any one of these files is useful. All four together is a self-improving workforce that runs while you sleep.

Subscribe now

How To Set This Up In 20 Minutes

This is not theory. Here is the build.

1. Open the Claude desktop app and switch to Cowork mode.

2. Pick a folder on your computer to be Claude’s home base. Mine sits at /Documents/Claude.

3. Inside that folder, create four files: Agents.md, Context.md, Memory.md, Skills.md.

4. Spend five minutes per file. Stub them in. They do not need to be perfect on day one.

5. Start a new chat in Cowork pointed at that folder. Tell Claude to read all four files and give you a one-paragraph summary of what it now knows about you.

That is it. You are operating, not prompting.

The files improve as you use them. Memory.md grows on its own. The other three you edit once a week, max.

My Agents In Action

The proof is in the time saved:

• Email Agent. Saves me about 1 hour a day.

• Brief Agent. Saves me about 30 minutes a day.

• Content Agent. Saves me about 30 minutes a day.

Two hours a day. Ten hours a week. Five hundred hours a year.

Once the file structure is set, you can schedule tasks to run your SOPs automatically. The Email Agent works while I am at the gym. The Brief Agent runs the night before every meeting. The Content Agent fills the gaps in my publishing calendar without me touching it.

I do not start the work anymore. I review it.

The Point

Stop treating Claude like a search engine.

Start treating it like a system.

Pick one file from this list. Build it this week. Five minutes is enough to start. The other three will feel inevitable once the first one is paying off.

The people winning the next five years are not the ones with the best prompts. They are the ones who built a system around Claude, not a habit of asking it questions.

Thanks for reading.

Much love ✌️

MJ

P.S. What workflows are you going to build? Hit reply and tell me. I read every one.

Most people use Claude like a fancy autocomplete.

Ask a question. Get an answer. Close the tab.

That’s not where the leverage is.

The leverage is in the commands. The little slash-prefixed shortcuts that turn Claude from a chatbot into an operator. I run five of them every day, and together they pull more than an hour back into my calendar.

Not in theory. In practice.

Here are the five I actually use, what they do, why they matter, and the exact prompts I run with each one. Copy them. Adapt them. Use them this week.

1. /skill-creator — The System

What it does: Builds a reusable Skill from any workflow I describe. Once it’s built, I run it anytime without re-explaining the process. It bakes in my voice, my rules, and my preferences.

Why it matters: My content strategy is built on repurposing. Every Substack issue should become multiple LinkedIn posts across my 70/20/10 content matrix (reach, trust, convert). This Skill automates the repurposing pipeline while respecting my voice rules. Build it once, use it every week.

The prompt:

/skill-creator Create a Skill that takes any Substack newsletter I've written and turns it into 3 LinkedIn posts: one list-style reach post, one personal story trust post, and one framework post with a lead magnet CTA. Use my writing rules (short sentences, no em dashes, no buzzwords, no AI voice). Include a hook, body, and CTA for each. Output as a ready-to-post draft.

2. /artifact — The Builder

What it does: Creates fully interactive mini-apps right inside Claude. Planners, calculators, dashboards. No code required.

Why it matters: I’m building a personal brand across LinkedIn, Substack, and Instagram (just started) simultaneously. A single dashboard that tracks content output across all three platforms means no more bouncing between Notion, spreadsheets, and analytics tools for a quick status check. One place. One source of truth. Open it, scan it, move on.

The prompt:

/artifact Build me a weekly content dashboard with three sections: (1) LinkedIn post tracker with columns for date, topic, pillar (Tech & AI, Systems Thinking, Personal Brand, Distribution), format (list/story/framework), status (draft/scheduled/posted), and impressions; (2) Substack pipeline with issue title, status, and publish date; (3) Instagram content tracker with post type, topic, and engagement. Make it interactive with filters and an "Add Post" button.

3. /chart — The Signal

What it does: Turns raw data into visual charts with instant analysis. I give Claude numbers. I get back a chart and a breakdown.

Why it matters: My personal brand is my distribution moat. Seeing follower growth and per-post impressions side by side reveals whether growth is coming from volume or virality. This is the kind of chart I drop into a Substack issue or use to pitch brand deals, without opening a spreadsheet.

Subscribe now

The prompt:

/chart My LinkedIn growth over the last 6 months: Oct 42K followers, Nov 46K, Dec 49K, Jan 53K, Feb 58K, Mar 64K. Also chart impressions per post: Oct avg 12K, Nov 18K, Dec 22K, Jan 31K, Feb 38K, Mar 50K. Show me the growth rate for both and when I'll hit 100K followers at this pace.

4. /ask — The Clarity

What it does: Makes Claude ask me clickable clarifying questions before it starts working. No fancy prompting required.

Why it matters: I post across multiple pillars and formats. Instead of getting a generic first draft, /ask forces Claude to nail down the specifics first. Is this a reach post or a trust post? Should it lead with the personal story or the tactical framework? The clickable options make it fast. I click a few buttons and the draft comes back dialed in.

The prompt:

/ask I want to write a LinkedIn post about how I use Claude Cowork to save 1-2 hours a day. Ask me about the format (list post, story, framework), which use cases to feature, the target audience, and the CTA before writing the draft.

5. /remember — The Memory

What it does: Saves my preferences, writing style, and background info to Claude’s memory so it remembers me across conversations.

Why it matters: Every conversation starts with context. No more repeating my voice rules, my banned phrases, or my role at Stanley every time I open a new chat. Claude sticks to one rule: if it sounds generic, it’s wrong. Loaded once, enforced every time.

The prompt:

/remember I'm MJ Jaindl, GM of Stanley (AI tool for LinkedIn content creation). I have 66k LinkedIn followers and 2k+ Substack subscribers. I also own Rock Star Crystals in NYC. My content pillars are Tech & AI Fluency, Systems Thinking, Personal Brand, and Distribution. I write in short, punchy sentences. No em dashes. No corporate buzzwords. No AI voice. Direct, specific, and personal. Never use "leverage," "synergy," "game-changer," or "Let me be honest."

The Bigger Picture

Here’s what most people get wrong about AI.

They use it like a search engine with better manners.

The real move is treating Claude like an operator. Give it a workflow (/skill-creator). Give it a surface (/artifact). Give it data (/chart). Give it constraints (/ask). Give it context (/remember).

That’s the difference between a chatbot and a system.

Each command on its own is useful. Together, they’re the reason I get an hour back every day.

Pick one this week. Set it up. Run it tomorrow.

Then add the next.

Thanks for reading. If this was useful, share it with someone who’s still copy-pasting the same context into Claude every morning.

Much love ✌️

MJ

That is 10% of what it can do.

I have been running Claude daily for over a year across Stanley, Rock Star Crystals (my crystal, gem, and mineral business), and my LinkedIn content engine. There are 9 distinct ways to use it. Most people only know one. Each mode unlocks a different kind of work, and the people who compound across all 9 are the ones pulling ahead right now.

This week I am breaking down all 9. For each one you get:

• What it actually is, in plain language

• How to turn it on

• A real use case you can steal

• A scaffolded prompt you can copy, fill in, and run today

• One pro tip most people miss

Pick one. Run it this week. Then come back for the next.

1. Chat

The conversation window. Where everyone starts, and where most people stop.

What it is

The chat window is the front door. One question, one answer, no files, no infrastructure. 99% of people never leave this mode, which is why 99% of people get chatbot-level results.

The upgrade inside Chat is Opus 4.7 (just released) with Extended Thinking. Opus is the flagship model. Extended Thinking tells it to reason before it answers instead of snapping out the first thing it thinks of. The difference is not subtle. A chat-style response is a guess. A thinking response is a plan.

How to turn it on

Open a new chat. Click the model selector at the bottom of the compose box. Pick Opus 4.7. Then toggle Extended Thinking on from the same menu. The toggle now uses adaptive effort levels (medium by default, bump to high or max for harder problems). If you do not see Opus, you are on a plan that does not include it. Upgrade to Pro, Max, or Team.

Use Case

You are prepping for a board meeting and need to pressure-test a strategic argument before you walk in. Chat is the right mode for this. One question, one thinking partner, no file creation, no infrastructure. Turn on Opus 4.6 with Extended Thinking and it will actually stress-test you instead of agreeing with you.

Prompt

You are a skeptical board member who has seen hundreds of pitches like mine.

Here is the argument I am planning to make: [PASTE ARGUMENT]

Context about the business: [1-2 SENTENCES]
What I am asking for: [DECISION / APPROVAL / BUDGET]

Do three things:
1. Name the 3 weakest assumptions in my argument.
2. Ask me the 3 hardest questions a real board member would ask.
3. Tell me what I should cut, add, or reorder before I walk in.

Be direct. Do not hedge.

Pro tip

Default Claude is too agreeable. Tell it to be skeptical, tell it who to roleplay as, and tell it not to hedge. The model obeys instructions about posture more than almost any other input. “Be direct. Do not hedge.” is the single most valuable line you can add to a Chat prompt.

2. Projects

Persistent workspaces. Conversations that do not forget.

What it is

A Project is a folder that Claude carries into every conversation. Drop in files, add instructions, and every new chat inside that Project starts with all of it already loaded. No re-explaining your voice. No re-uploading your style guide. No copy-pasting the same context every Monday.

As of March 2026, Projects now work inside Cowork too, with persistent memory, scheduled tasks, a context folder, and custom instructions scoped to that one project. If you were using Projects a year ago, they are three times more capable now.

How to turn it on

In the Claude app sidebar, click Projects, then New Project. Give it a name. Add files under Project Knowledge (PDFs, docs, markdown, whatever). Add Custom Instructions that tell Claude how to behave inside the project. Every new chat you start from inside that project inherits all of it automatically.

Use Case

You keep re-explaining your brand, your customers, your goals, and your voice every time you open a new chat. Stop. Create a Project called Personal Brand, drop in your voice guide, audience doc, and content pillars once, and every conversation inside that project inherits all of it. Same for a company. Same for a client.

Prompt

(Inside a Project that already has your context loaded)

Act as my content strategist. You already have my voice guide, audience, and pillars in this project.

I want to plan next month's content. Give me:
- 8 post ideas mapped to my pillars (2 per pillar)
- The hook line for each
- Which format fits best (listicle, story, breakdown)
- Which 2 are most likely to go viral and why

Pull directly from my voice guide. Do not invent a new tone.

Pro tip

Custom Instructions beat Project files for behavior. Files are what Claude knows. Custom Instructions are how Claude acts. Put your voice rules, banned words, and output format in Custom Instructions, not in a markdown file. You will feel the difference in the first reply.

3. Skills

Saved playbooks. Your repeatable workflows on autopilot.

What it is

A Skill is a reusable workflow Claude can run on command. You write it once as a SKILL.md file with a short description and a set of instructions, and Claude now has a new tool. Trigger it with a slash command like /linkedin or let Claude auto-invoke it when it recognizes the task. Skills can bundle multiple files, call scripts, and orchestrate multi-step work. This is how power users stop prompting and start shipping.

How to turn it on

The fastest path is to ask Claude to build the skill for you. Tell it what task you repeat, what inputs you provide, and what output you want. It will generate a SKILL.md file with the right YAML frontmatter. Save it in your skills folder. From that point on, /your-skill-name runs the whole playbook.

Use Case

You write a email update every weekday. The process is the same every time: rough idea, structure it, match your voice, tighten it, add the hook. That is a Skill. Build it once with a slash command like /update and now every post starts from a 90-second scaffold instead of a blank page. Same principle for weekly reports, client updates, sales follow-ups, anything you repeat.

Prompt

Help me build a Skill for [REPEATABLE TASK, e.g., "writing my email update"].

Inputs I will provide each time: [E.G. ROUGH IDEA, TARGET AUDIENCE, GOAL OF POST]

The Skill should:
1. Ask me any missing inputs before writing anything.
2. Run the task using [MY VOICE GUIDE / TEMPLATE / RULES - reference the file].
3. Return the output in [FORMAT, E.G. DRAFT].
4. End with 2 alternate hooks I can swap in.

Write the skill as a reusable prompt I can save and trigger with one command.

Pro tip

The description field in the YAML frontmatter is everything. That is what Claude reads to decide whether to auto-run the skill. Write it like a Google search query, not a sentence. “Use this skill when the user asks about X, Y, or Z” beats “This skill helps with content creation” every time.

4. Cowork

Your desktop partner. The mode that ships real files.

What it is

Cowork is the mode that opens your folders, reads your files, and saves real deliverables back to your computer. Word docs with working heading styles. Excel files with live formulas. PowerPoint decks. PDFs. Not a preview. Not a text block. The actual file, in the folder you pointed to. It runs on macOS and now Windows (since February 10, 2026) and it is the setting I spend the most time in.

How to turn it on

Download the Claude desktop app. Open it and switch the mode selector to Cowork. At the bottom of the compose box, check “Work in a Folder” and pick a folder from your computer. From that point on, Claude reads and writes into that folder. Give it a task and walk away.

Use Case

You finish a meeting, have 40 screenshots named IMG_4782.png, three half-written SOPs, and a Notion to-do list you have not touched in a week. Cowork opens your actual folders, reads the files, renames the screenshots, updates the SOPs in place, and pushes the action items into Notion while you move on with your day. It does not hand you a summary to copy-paste. It does the work.

Prompt

You have access to my [FOLDER NAME] folder and my Notion.

Do the following in order:
1. Read every file in [FOLDER] modified in the last 7 days.
2. Rename any files with generic names (IMG_xxxx, Untitled, etc.) based on their content.
3. Identify any SOPs or docs that are incomplete or out of date. List them.
4. For each to-do or action item you find buried in those files, add it to my Notion to-do list under [DATABASE NAME] with a due date if one is mentioned.

Tell me what you did when you finish. Do not ask permission for renames or Notion updates.

Pro tip

Cowork + a Project is the real setup. Wrap Cowork in a project so every task inherits your voice, your folder structure, and your rules. Once you do this, you are not prompting anymore. You are assigning work.

5. Code

Write software by describing what you want.

What it is

Claude Code is a coding agent that lives in your terminal and ships software. It reads your codebase, writes features, runs tests, commits to GitHub, and opens pull requests. As of April 14, 2026, Routines lets it run on a schedule, react to GitHub PRs, or trigger from an API call. This is not autocomplete. This is a junior engineer that works while you sleep.

How to turn it on

If you are a builder, install Claude Code from claude.com/product/claude-code and run it in your terminal. If you are not a developer, use the Code mode inside the Claude desktop app. Same engine, no terminal. Describe the feature, watch it build.

Use Case

You are not a developer but you need a simple internal tool. A form that collects customer data and drops it into a Google Sheet. A landing page for a new offer. A script that pulls last week’s sales from Shopify and emails you a summary. Claude Code writes it, pushes it to GitHub, and can ship it live without you touching a terminal.

Prompt

I want to build [TOOL / FEATURE / SCRIPT] and I do not want to write the code myself.

What it needs to do:
- [FUNCTION 1]
- [FUNCTION 2]
- [FUNCTION 3]

Constraints:
- Language / framework: [OR "YOU CHOOSE - PICK THE SIMPLEST OPTION"]
- Where it runs: [LOCAL / WEB / CLOUD / INSIDE APP X]
- Who uses it: [ME ONLY / MY TEAM / CUSTOMERS]

Build it. Then:
1. Explain what each file does in one sentence.
2. Tell me exactly how to run it.
3. Flag anything I need to set up first (API keys, accounts, installs).

Pro tip

Always ask it to write tests first. “Write failing tests for this feature, then make them pass.” You will catch bugs before they ship and you will get a working test suite for free. This single instruction upgrades the quality of the output by 40%.

6. Computer Use

Hand Claude the mouse.

What it is

Computer Use is screen takeover. Claude sees your screen, moves your cursor, clicks buttons, and types into forms. It is built for cross-app jobs that no single integration can solve. Launched March 23, 2026 as a research preview for Claude Pro and Max users. macOS only today, Windows coming. Dispatch lets you send a task from your phone and come back to a finished job.

How to turn it on

Open the Claude desktop app. Switch to Computer Use mode. Grant screen recording and accessibility permissions the first time. Give it a task. It takes over the mouse and keyboard and narrates what it is doing as it goes. You can stop it at any time.

Use Case

You need to fill out the same onboarding form for 15 new vendors, each in a different web portal. Or you need to open 10 PDFs, grab a specific field from each, and paste it into a spreadsheet. Computer Use opens the apps, clicks through the menus, types into the fields, and does it on your behalf. Send the task from your phone in the morning. It is done when you get back to your desk.

Prompt

I need you to do the following on my computer.

Task: [CLEARLY STATE THE END GOAL]

Steps I want you to follow:
1. Open [APP / WEBSITE]
2. For each [ITEM], do [ACTION]
3. Save the result to [LOCATION / FILE]

Before you start:
- Confirm you have access to the apps you need.
- Tell me if anything is ambiguous.
- Do not click any links or submit any forms that send money, send email to external contacts, or change account settings without asking me first.

When you finish, show me what you did.

Pro tip

Always include a “do not do” list. Computer Use will do exactly what you tell it to, including things you did not mean to say. “Do not send money, do not email external contacts, do not change settings” is not paranoia. It is the instruction that keeps a 10-minute task from turning into a cleanup project.

Subscribe now

7. Connectors

Plug Claude into the tools you already use.

What it is

Connectors are one-click integrations with the apps your work actually lives in. Gmail, Google Calendar, Google Drive, Slack, Notion, HubSpot, Stripe, Canva, and 50+ others. Free with your subscription. Once connected, Claude can read from and write to those tools directly, which means no more copy-pasting context between windows.

How to turn it on

Go to Settings, then Connectors, then Browse. Find the tool. Click Add. You will be redirected to the tool’s OAuth screen, approve access, and you are done. The whole thing takes 30 seconds per tool. Connect Gmail, Calendar, Google Drive, Notion, and Slack first. That gets you 80% of the value.

Use Case

Your real work lives in Gmail, Google Drive, Notion, Slack, Granola, and Calendar. Connectors let Claude read and write directly to those tools instead of you copying and pasting between them. The second you connect Gmail, you can ask Claude what needs a reply today and it will actually tell you. Setup takes seconds. Payoff is permanent.

Prompt

(After connecting Gmail, Calendar, and Notion)

Act as my assistant for the next 10 minutes.

Do this in order:
1. Scan my Gmail inbox from the last 48 hours. Surface only emails that need a reply from me, grouped by urgency (today / this week / can wait).
2. Check my calendar for tomorrow. Flag any meeting I do not have prep notes for.
3. For each flagged meeting, create a prep doc in Notion under [DATABASE] with attendees, context from any related emails, and 3 questions I should be ready to answer.

Report back in a short summary, not a wall of text.

Pro tip

Connect one tool at a time and actually use it for a week before adding the next one. If you connect 10 at once, Claude has too many options and picks the wrong one. If you connect Gmail, use it hard for a week, then add Calendar, the workflows compound and actually stick.

8. Excel

AI that actually understands your spreadsheet.

What it is

Claude for Excel is a Microsoft 365 add-in that installs inside Excel. Not a web interface that reads your file. The real thing, inside the app, with full awareness of every tab, every formula, and every cell reference. Claude can trace dependencies, catch broken formulas, explain what a model is doing, and rewrite cells while preserving the logic underneath. Available on Pro, Max, Team, and Enterprise plans.

How to turn it on

Go to the Microsoft Marketplace, search “Claude for Excel,” click Get It Now. The add-in installs into Excel. Once it loads, hit Control+Option+C on Mac or Control+Alt+C on Windows to open the side panel. Sign in with your Claude account. You are in.

Use Case

You inherited a 2,000-row spreadsheet from someone who left the company. Half the formulas are broken, the logic is undocumented, and you need to present from it in two days. Claude reads every tab, explains what the model is doing, finds the broken references, and writes the formulas you are missing. It treats your workbook like a system, not a grid of cells.

Prompt

Open [FILE NAME] and walk me through it like I have never seen it before.

1. Tell me what this workbook is for in one sentence.
2. List every tab and what each one does.
3. Identify any formulas that are broken, circular, or hardcoded where they should not be.
4. Tell me the 3 cells that matter most (the ones driving the final output).
5. If I wanted to [ADD / CHANGE / REMOVE X], which cells and tabs would I need to touch?

Then wait for my next instruction.

Pro tip

Ask for cell-level citations. When Claude explains a formula or suggests a change, tell it to cite the exact cell reference (”B14,” not “the revenue cell”). It forces the model to check the workbook instead of pattern-matching what a spreadsheet usually looks like, and you can verify the answer in two clicks.

9. Artifacts

Instant apps and functional tools generated mid-conversation.

What it is

An Artifact is a live, interactive thing that Claude builds inside the chat window. A calculator. A dashboard. A quiz. A dynamic chart. A working web app. Not a screenshot of one. The real thing, rendered live, that you can click through, edit, and share. Claude builds them as React components with Tailwind styling by default, so they look good out of the box.

How to turn it on

Nothing to turn on. Ask for something interactive and an Artifact appears in a side panel next to the chat. On the free plan you now get Artifacts out of the box. Publish it with one click to get a shareable link you can send to a prospect or embed on your site.

Use Case

You are explaining pricing to a prospect and realize a calculator would land way better than a spreadsheet. Ask Claude to build a pricing calculator as an Artifact. Thirty seconds later you have a working, interactive tool inside the chat window you can share with them. Same for dashboards, quizzes, content planners, lead magnets. Outputs you interact with, not just read.

Prompt

Build me an interactive [TOOL TYPE, E.G. "PRICING CALCULATOR"] as an Artifact.

What it should do:
- Inputs: [LIST INPUTS, E.G. "NUMBER OF USERS, PLAN TIER, ADD-ONS"]
- Calculation: [LOGIC, E.G. "BASE PRICE + PER-USER + ADD-ONS, WITH ANNUAL DISCOUNT"]
- Output: [WHAT THE USER SEES, E.G. "TOTAL MONTHLY AND ANNUAL COST, PLUS SAVINGS"]

Design:
- Clean, minimal, mobile-friendly.
- Match this vibe: [LINK, BRAND DESCRIPTOR, OR "KEEP IT DEFAULT"]

Make it work end-to-end. Do not stub out values.

Pro tip

Iterate inside the artifact, not the chat. When something is wrong, do not re-paste the whole prompt. Point at the specific element (”the submit button is misaligned,” “the discount logic is off by 10%”) and let Claude patch the live version. You keep the working parts and only change what needs changing. Five rounds of this and you have a polished tool.

The Point

Most people will read this, nod, and keep using Claude the exact same way they did yesterday.

Do not be that person.

Pick one mode on this list you have never used. Copy the prompt. Fill in the brackets. Run it today. Let the result tell you which mode to try next.

The people who will win the next five years are not the ones with the best prompts. They are the ones who treat Claude as a system with 9 entry points, not a search bar with one.

Thanks for reading.

Much love ✌️

MJ

Every morning I used to start the same way: check my calendar, scan emails, open Notion for my task list, review my content calendar for gaps. Four apps, 20-30 minutes, before I’d even started real work.

Now I wake up to a single document that has all of it done for me. Meetings with context on what we last discussed, emails filtered to only what matters, project tasks organized by priority, and a two-week content calendar that flags exactly which days are missing posts.

Here’s exactly how I built it using Claude Cowork, step by step, so you can build your own.

Part 1: Setting Up the Connectors

The briefing pulls live data from four apps. Each one is a “connector” in Cowork. Basically a permission that lets Claude read from that app on your behalf. Here’s how to turn them on.

Google Calendar

This is what feeds the “Today’s Meetings” section. In Cowork, go to Settings → Connectors and enable Google Calendar. You’ll authorize with your Google account, and Claude gets read access to your primary calendar.

What it does in the briefing: Pulls today’s events with times, titles, and attendees.

Gmail

Same process. Enable the Gmail connector in Settings. This gives Claude read access to your inbox.

What it does in the briefing: Searches for unread emails from the last 24 hours, filters out promotions and spam, and surfaces anything that needs action.

Notion

Enable the Notion connector. You’ll authorize through Notion’s integration flow and select which workspaces/pages Claude can access.

What it does in the briefing: Pulls from two Notion databases: my Project Tasks database and my Content Calendar database.

Granola

Enable the Granola connector. Granola is a meeting notes tool that auto-records and transcribes your calls.

What it does in the briefing: For each meeting on today’s calendar, Claude looks up the last few conversations with those attendees and summarizes what you discussed and what the open action items are. This is the part that consistently surprises me. I walk into every meeting with full context.

Total setup time: about 5 minutes. You authorize each one once and they stay connected.

Part 2: Creating the Scheduled Task

Once the connectors are live, you create a “Scheduled Task” in Cowork. This is essentially a prompt that Claude runs automatically on a schedule.

Here’s what I did:

1. Opened a new Cowork session and told Claude what I wanted: a daily briefing pulling from calendar, email, Notion tasks, and my content calendar.

2. Claude created the scheduled task with these settings:

   • Task name: daily-morning-briefing

   • Schedule: Weekdays at 7:00 AM ET (cron: 0 7 * * 1-5)

   • Prompt: The full set of instructions (we’ll break this down next)

3. Ran a test immediately. This is important. The first run prompts you to approve tool permissions (Google Calendar, Gmail, Notion, Granola). Those approvals carry forward to all future automatic runs. If you skip this, your first scheduled run might pause waiting for permissions.

That’s it for setup. Now let’s get into the actual prompt. That’s where the real magic happens.

Part 3: The Prompt, Section by Section

I’m going to show you each section of the prompt I gave Claude, then show you what that section actually produces in the briefing. The prompt is long and specific. I’ve learned that the more precise you are, the better the output.

Section 1: Today’s Calendar + Meeting Context

The prompt:

Use the Google Calendar tool to pull today’s events from the primary calendar. Timezone: America/New_York. List each event with time, title, and attendees. Then, pull my last few Slack and Granola conversations with the attendees and summarize in 3-5 bullets what we talked about and our action items.

Why I wrote it this way: The calendar pull is straightforward, but the second part is what makes this valuable. By telling Claude to cross-reference attendees against Granola meeting notes, I get a prep brief for every meeting. Instead of walking in cold, I know exactly what we discussed last time and what’s still open.

What the output looks like:

Subscribe now

Section 2: Email Highlights

The prompt:

Use the Gmail tool to search for unread emails from the last 24 hours: query “is:unread newer_than:1d”, maxResults 15. Filter out obvious promotions/spam. Surface emails from the Stan founder John Hu or Vitalii Dodonov. Surface emails related to deals. Surface emails that request something from me. Flag anything with a deadline or action required.

Why I wrote it this way: I don’t need to see every email. I need to see emails from specific people (my key business contacts), deal-related emails, and anything where someone is waiting on me. Everything else can wait. The “flag anything with a deadline” instruction catches things like compliance deadlines or Ramp receipts that have due dates.

Section 3: Notion Project Tasks

The prompt:

Use the Notion tool to search the Project Tasks database for active tasks. Query: “tasks not started or in progress.” Fetch the top results and categorize them by area: Stanley/Business, Rock Star Crystals, Content/Growth, Personal/Financial. Highlight any that are overdue or high priority.

Why I wrote it this way: I manage three businesses and personal priorities all in one Notion database. Without the categorization instruction, Claude would just dump a flat list. By telling it to group by area, I can quickly see what’s urgent at Stanley vs. what’s personal. The “highlight overdue or high priority” instruction makes sure I don’t miss deadlines buried in a long list.

What the output looks like:

Section 4: Content Calendar with Gap Detection

This is the most complex section of the prompt, and the one that took the most iteration to get right. Here’s why.

The problem I ran into: Notion’s search is semantic, not structured. When you search for “posts this week,” it returns results based on text relevance, not actual dates. So Claude would confidently tell me “Wednesday is open, needs content” when there was actually a post scheduled for Wednesday that just didn’t match the search terms.

The solution: I wrote a multi-pass search strategy. Here’s the prompt:

Run at least 7 separate Notion searches against the Content Calendar database with page_size=25 each. Use DIFFERENT broad queries each time to maximize coverage:

• “LinkedIn post”

• “newsletter article”

• “carousel image text”

• “growth AI automation”

• “personal brand content creation”

• “tools system strategy”

• “draft ready backlog scheduled posted”

Also run searches with date range filters starting from 2-3 weeks ago. Collect all unique page IDs and deduplicate. Then fetch EVERY unique page to read its structured properties: Name, Date, Status, Post Type, Platform, Content Pillars, Visuals needed.

The “date:Date:start” property is the ONLY reliable way to know what date a post is assigned to. Never infer dates from search result timestamps or page titles.

Why this matters: This multi-pass approach casts a wide net. Different search terms surface different posts, and deduplicating by page ID means nothing gets counted twice. Then by fetching each page individually, Claude reads the actual structured data (dates, status fields) instead of guessing from search snippets.

I also defined my posting schedule in the prompt:

Monday through Friday: 1 LinkedIn post per day. Saturday: 1 Newsletter post. Sunday: No post.

And gave Claude specific rules for how to display statuses:

“Posted” → ✅ Posted. “Scheduled” → ✅ Scheduled. “Draft” → ⚠️ Draft (needs finishing). “Backlog” → ⚠️ Backlog (idea stage). If a day has no matching entry after exhaustive search, mark it as “No post found (verify in Notion)” rather than definitively declaring it missing.

That last line is key. I’d rather have the briefing say “I’m not sure, go check” than confidently tell me a day is empty when it’s not.

What the output looks like:

Section 5: Compile and Prioritize

The prompt:

Present the briefing in this order: Calendar, Project Tasks, Content Calendar, Email Highlights, Action Items. Top 5-6 suggested priorities for the day. Keep the tone concise and professional. Use tables for the content calendar. Link to Notion pages where possible.

Why I wrote it this way: The ordering is intentional. Calendar first because meetings are time-sensitive. Tasks and content calendar next because that’s where I spend most of my day. Email last because it’s reactive, not proactive. And the “Action Items” section at the end gives me a prioritized punch list I can work through without thinking.

What I’ve Learned After a Few Weeks

Be absurdly specific. Vague prompts get vague results. The more explicit you are about what to search, how to categorize, and how to display, the more useful the output.

Semantic search needs workarounds. The content calendar section went through three iterations before I figured out the multi-pass approach. If you’re pulling from a database, don’t trust a single search to find everything.

Test, then iterate. My first version of this prompt was half as long. Every time the briefing missed something or got something wrong, I added a line to the prompt to fix it. The prompt is a living document.

The meeting context is the killer feature. I didn’t expect this when I started, but having Granola meeting notes summarized alongside today’s calendar saves me more time than any other section.

If you’re using Cowork (or any AI tool with app integrations), I’d encourage you to try building something like this. Start simple. Even just calendar + email. Then add complexity as you learn what you actually need every morning.

The goal isn’t to automate everything. It’s to make sure you never walk into your day blind.

Happy Saturday

MJ ✌️

So I set up OpenClaw on a Mac Mini M4.

And now I have an always-on AI agent I can message from Discord, anywhere in the world.

Here’s the full setup, start to finish.

Why Mac Mini + OpenClaw

OpenClaw is an open-source AI agent platform. It connects to Claude (for reasoning) and Perplexity (for search), then layers on skills like Gmail, Google Calendar, Notion, and more.

The Mac Mini M4 is the perfect host. Low power draw, Apple Silicon handles AI workloads well, and after initial setup, you don’t need a monitor. (after it’s all set up that is) It just sits there and works.

Total cost: the Mac Mini itself, plus pay-as-you-go API fees for Claude (these can get a little pricey; I use Claude Haiku for most tasks because it’s cheaper) and Perplexity. That’s it.

Let’s get into it.

Step 1: Create a New Gmail Account

Don’t use your personal Gmail. OpenClaw needs a dedicated Google account to access Workspace services without mixing in your personal data. A separate account keeps everything clean and easy to revoke later if needed.

Go to accounts.google.com, click “Create account,” and pick a name that reflects its purpose (something like mj-openclaw@gmail.com). Skip phone verification for now. You can add it later for recovery.

Step 2: Create a New Apple ID

Same logic. You want a clean Apple ID that isn’t tied to your personal iCloud, purchases, or Find My network.

Go to appleid.apple.com, click “Create Your Apple Account,” and use the new Gmail you just made as the Apple ID email. Set a strong password and store it somewhere safe. Password manager recommended.

Step 3: Start Up Your Mac Mini

Unbox it. Connect power, a monitor (just for setup), keyboard, and mouse. Walk through the macOS setup: language, Wi-Fi, terms.

After this initial setup, you can run it headless. No monitor needed. It just sits on a shelf and does its thing.

Step 4: Log In with Your New Apple ID

During setup (or in System Settings after), sign in with the new Apple ID you just created. Skip iCloud, Find My, and all the extras. Create a local user with a strong password.

Using a dedicated Apple ID keeps this machine completely separate from your personal setup. Better security, cleaner separation.

Step 5: Open Terminal and Install OpenClaw

This is where it gets real. One command installs OpenClaw and sets it up as a background service.

Open Terminal (search for it in Spotlight, or find it in Applications > Utilities). Run:

curl -fsSL <https://openclaw.ai/install.sh> | bash

When the install finishes, run:

openclaw onboard --install-daemon

The onboarding wizard walks you through gateway settings. Accept the defaults unless you have a reason to change them. That’s it. OpenClaw is installed and running as a background service on startup.

Step 6: Open Your Browser to Finish Setup

After the CLI onboard completes, OpenClaw launches a local web dashboard. This is your control center.

Open any browser and go to:

http://127.0.0.1:18789/

This is the OpenClaw dashboard. Your hub for agents, skills, and services. Make sure the gateway status shows “Running” before you move on.

Step 7: Set Up Messaging with Discord

This is the part that makes it feel like magic. Discord becomes your remote control for OpenClaw. You can message your agent from your phone, your laptop, anywhere.

Create a Discord server for your agent. Install the Discord skill in OpenClaw and connect it via bot token. Message your agent to run tasks. Use a channel like #openclaw-tasks to stay organized.

Now you’re not tethered to the Mac Mini. You’re sending tasks from your phone while walking the dog.

Step 8: Get Your API Keys for Anthropic and Perplexity

OpenClaw needs two API keys to function. Claude handles reasoning. Perplexity handles search. Both are pay-as-you-go.

Anthropic API Key: Go to console.anthropic.com. Create an API key. Paste it into OpenClaw. Add a payment method.

Perplexity API Key: Go to perplexity.ai. Generate an API key. Save it for OpenClaw.

Paste both keys into the OpenClaw dashboard. Done. Your agent can now think and search.

Step 9: Connect Your Skills

Skills let OpenClaw actually use your tools. This is where it goes from “cool setup” to “this thing is doing real work for me.”

Go to Skills in the dashboard and install:

Google Workspace: Install the “gog” skill and sign in. This gives OpenClaw access to Gmail, Calendar, and Drive.

Notion: Install and authorize access. Now your agent can read and update your Notion workspace. I currently use better-notion.

Perplexity: Install and paste your API key. Web search, unlocked.

Anthropic: Install and paste your API key. Claude-powered reasoning, connected.

Add whatever else matches your workflow. The more skills you connect, the more your agent can do without you.

You’re Live

Your Mac Mini now runs 24/7 as your personal AI agent.

Message it from Discord. Have it check your email, update your tasks, search the web, manage your calendar. It doesn’t sleep. It doesn’t forget. It just runs.

Total setup time: about an hour or two if you’re taking it slow. The hardware costs what a Mac Mini costs. The API fees are pennies per task.

One hour of setup. Unlimited hours saved.

That’s the trade.

P.S. If you set this up, message me on LinkedIn and tell me what you’re using it for. I want to hear it. https://www.linkedin.com/in/jaindl/

Much love ✌️

MJ

Subscribe now

The Democratization Problem

Here is what McKinsey’s latest data shows: 88% of organizations now use AI. 79% use generative AI. These numbers went from 33% to 79% in just two years. That is not adoption. That is commoditization. AI is now a utility, not an advantage. Anyone can wire systems together. Anyone can deploy agents. The technical skill barrier is gone.

McKinsey AI Adoption Timeline Source: McKinsey “AI at work but not at scale”

But here is what happened next: Only 6% of companies are actually capturing meaningful value from AI. The other 94% are stuck experimenting and spending money without returns. Why? Because technical ability is no longer scarce. What is scarce now is trust, attention, and distribution.

Subscribe now

Meet the New Arbitrage: Technical Plus Brand

The new winners in the AI era aren’t the best engineers. They’re the best engineers with a personal brand. Look at the data: 76.4 million Americans freelance. But 4.7 million earn over 100K per year. That’s 6%. What separates them from everyone else? Personal brand and distribution.

78% of CEOs say their top freelancers contribute more value than salaried employees with degrees. But those top freelancers? They’re not hidden. They’re visible. They have a following. They market themselves. Here’s the real number that matters: AI-related freelance projects jumped 60% year-over-year. And who’s winning? The 54% of freelancers with advanced AI skills who also have a public presence.

AI Freelancer Rate Premium Zen van Riel AI Engineer Rates

The silent expert? They’re making the same as everyone else. Oversaturated. Forgotten. The expert with a 50K LinkedIn following? They’re 10x ahead. Inbound clients. Zero cold outreach needed.

The Contract Economy Is Already Here

This isn’t coming. It’s happening now. By 2028, over 50% of the U.S. workforce will be freelance. That’s not some distant future. That’s three years away. Companies are shrinking. Teams of 2-3 people running AI agents are doing what used to take 20. They’re faster. They’re cheaper. They’re leaner.

U.S. Freelance Workforce Growth Source: Carry.com Freelancing Statistics (data from Upwork/Statista)

And those 2-3 people? They’re not anonymous. They’re building in public. They’re documenting their process. They’re becoming known in their space. The contract economy rewards visibility. 1.9 trillion dollars globally was spent on independent contractors in 2024. 65% of high-growth companies now hire freelancers regularly. And 45% of those companies pair freelancers with AI tools.

Freelance Platforms Market Size Source: Grand View Research

Translation: Small teams plus AI automation plus personal brand equals unstoppable.

What This Means For You

If you’re technical and thinking about staying quiet, being really good at your craft, and letting your work speak for itself: That strategy is already dead. The new formula is:

1. Build real AI systems (table stakes — everyone can do it).

2. Document the process (share what you learn, the mistakes you made, the insights you found).

3. Build a public audience (LinkedIn, newsletter, Twitter, YouTube — pick one and go deep).

4. Attract inbound opportunities (clients, partners, employees find YOU).

5. Monetize multiple ways (freelance work, consulting, digital products, coaching, equity partnerships).

The person who can wire systems together AND tell the story of wiring systems together wins. Not because they’re better at building. But because distribution is now the moat.

The Real Shift

AI didn’t create the scarcity of jobs. It created the scarcity of attention. Everyone can build. But not everyone can build AND communicate what they built to 50,000 people. The winners in 2026 won’t be the smartest engineers. They’ll be the smartest engineers who learned how to market themselves.

Your next customer isn’t on LinkedIn looking for “freelance AI developer.” They’re watching someone they know, like, and trust build in public. And that’s where you come in.

Build in public.
Share the lessons.
Attract the right people.
Win.

Happy Sunday ✌️
MJ

Most people use Claude like a chatbot.

They open a new conversation, type a prompt, get a response, and move on. Maybe they tweak the prompt a few times. Maybe they copy-paste the output into a Google Doc and spend 30 minutes editing it to sound like them.

That’s like buying a sports car and never leaving the parking lot.

Here’s the thing -the difference between generic AI output and output that actually sounds like you isn’t about writing better prompts. It’s about a better setup. One that takes about 30 minutes, runs once, and compounds every single session after that.

I’ve been using Claude’s Cowork mode daily for months now -for LinkedIn content, strategy docs, business planning, and research. The outputs I get today are unrecognizable compared to what I was getting when I first started. And it’s not because I got better at prompting. It’s because I dialed in these four steps.

Let me walk you through exactly what I do.

Step 1: Initial Configurations

Before you write a single prompt, you need to set four things inside the Claude desktop app. These are your foundation. Get them right once, and everything that follows works better.

Enable Extended Thinking.

This is the single biggest unlock most people skip. Extended Thinking lets Claude reason through problems step-by-step before giving you an answer -kind of like how you’d work through a complex problem on a whiteboard before presenting your conclusion.

Without Extended Thinking, Claude gives you its first instinct. With it, Claude actually thinks. It plans, reconsiders, catches its own mistakes, and produces significantly better output -especially for anything involving strategy, writing, analysis, or multi-step tasks.

You’ll find the toggle right in the model selection dropdown. Turn it on and leave it on.

Select Opus 4.6.

Not all Claude models are created equal. Opus 4.6 is Anthropic’s most capable model -it’s the one built for complex, ambitious work. It reasons more deeply, handles nuance better, and produces higher-quality writing than the lighter models.

Yes, it’s a bit slower. Yes, it uses more of your message quota. But if you’re using Claude for real work -not just quick questions -Opus is worth it every time. Think of it this way: you wouldn’t use a calculator app when you need a spreadsheet. Use the tool that matches the job.

Subscribe now

Use Cowork, not Chat.

This is the one that changes everything.

Most people use Claude through the Chat tab. That’s fine for quick questions and brainstorming. But Cowork is a completely different experience. In Cowork mode, Claude can actually read and write files on your computer. It can create documents, edit existing ones, organize folders, build presentations, and work with your real files -not just generate text in a chat bubble.

Think of Chat as texting your assistant. Cowork is like sitting next to them at a desk with all your files spread out.

When you’re in Cowork mode, Claude can create Word docs, spreadsheets, presentations, and markdown files. It can read documents you’ve already written and use them as context. It can iterate on work across multiple sessions. That’s the kind of capability that turns Claude from a novelty into a legitimate work tool.

Connect your working folder upfront.

Cowork needs a folder on your computer to work with. This is the folder Claude will read from and write to -your shared workspace.

Here’s my recommendation: create this folder inside a cloud storage service like Dropbox, Google Drive, or iCloud. That way the folder lives locally on your machine (which Cowork requires), but it’s also synced and backed up automatically. You get the best of both worlds -Claude can access it instantly, and you never lose your work.

Connect this folder the moment you open Cowork, before you start your first task. It takes five seconds and it means Claude has access to everything it needs from the jump.

Step 2: Create Your Folder Structure

Now that Claude is configured, you need to give it a workspace that’s organized. This isn’t busywork -it’s the foundation that makes every future session faster and better.

I keep my structure simple. Four folders. That’s it.

Create a main folder called “Claude Cowork.”

This is your home base. Everything Claude-related lives here. I keep it in my Dropbox so it syncs across devices and stays backed up, but any cloud folder works.

Create a subfolder called “Me.”

This is where your personal context files live -the files that tell Claude who you are, what you do, and how you work. More on this in Step 3, because this folder is where the magic happens.

Create a subfolder called “Strategy.”

This is where you store strategy docs, business plans, content calendars, competitive analyses -anything Claude helps you create that’s strategic in nature. Over time, this folder becomes incredibly valuable because Claude can reference previous strategy work when helping you with new projects. Context builds on context.

Create a subfolder called “Examples.”

This is where you put writing samples, templates, past deliverables, and reference material. If you want Claude to write like you, it needs to see how you write. If you want it to format something a certain way, show it an example. This folder is your “show, don’t tell” library.

The beauty of this structure is its simplicity. Three subfolders, each with a clear purpose. You’re not building a complex file system -you’re giving Claude just enough organization to find what it needs without drowning in noise.

Every word loaded into Claude’s context costs it “thinking room.” One bloated folder with hundreds of random files means Claude wastes energy sorting through noise. But too many nested folders means important files get overlooked. This three-folder setup hits the sweet spot.

Step 3: Create Your Me.md File

This is the step that separates “pretty good” Claude outputs from “how did it know exactly what I wanted?”

The Me.md file lives in your “Me” folder. It’s a simple markdown file -just a plain text file with some basic formatting that any text editor can create. But don’t let its simplicity fool you. This file is essentially your briefing document. Every time Claude starts a session, it reads this file and suddenly knows who it’s working with.

I structure my Me.md with three sections:

Section 1: My role, objectives, and daily tasks.

This is the “who I am and what I do” section. Be specific. Don’t just write “I’m a marketing professional.” Write what you actually do day-to-day.

For example, mine covers my role as GM of Stanley, my work as a LinkedIn content creator, and my business Rock Star Crystals. It includes my key objectives, the types of tasks I handle daily, and the outcomes I’m working toward.

The more specific you are here, the better Claude calibrates its responses. When Claude knows you’re a GM who also creates LinkedIn content, it won’t give you generic advice -it’ll tailor everything to your actual situation. Need a LinkedIn post? It’ll write it in the context of your brand. Need a strategy doc? It’ll factor in the realities of your role.

Subscribe now

Section 2: Writing samples.

This is where you paste examples of your actual writing. LinkedIn posts that performed well. Emails you’re proud of. Paragraphs from documents that capture your voice.

Claude is remarkably good at picking up on voice and style when you give it examples. Without samples, it writes in “generic AI voice.” With three to five strong samples, it starts writing in your voice. The more variety you include (casual posts, professional emails, long-form pieces), the better Claude understands the range of your tone.

Section 3: A list of “don’t dos” when writing.

This is the secret weapon. Everyone has writing tics they hate -things AI tends to do that make you cringe. Maybe you hate when Claude uses “dive into” or “leverage” or “in today’s fast-paced world.” Maybe you don’t want bullet points in your LinkedIn posts. Maybe you never use emojis.

Write it all down. Be specific and blunt. “Never start a post with a question.” “Don’t use the word ‘delve.’” “No corporate jargon.” “Don’t use more than one exclamation point per post.”

This negative space is just as important as the positive instructions. It tells Claude what to avoid, which dramatically reduces the editing you need to do after the fact. Instead of generating something and then spending 15 minutes stripping out the AI-isms, Claude avoids them in the first place.

Step 4: Set Global Instructions

The last piece of the puzzle is Global Instructions. These are persistent rules that apply to every single Cowork session -you set them once and they run in the background forever. No more repeating yourself at the start of every conversation.

You’ll find this in Settings, under Cowork, then Global Instructions. It’s a text field where you can write anything you want Claude to always do.

Here are the four instructions I use:

“Review my Me.md file before starting any task.”

This ensures Claude always reads your context file before doing anything. Without this, Claude might skip the file and give you generic output. With this instruction, it checks in with your Me.md every single time -which means every response is calibrated to your role, your voice, and your preferences.

“Ask clarifying questions before executing.”

This one saves enormous amounts of time. Without it, Claude will take your prompt at face value and run with it -sometimes in a direction you didn’t intend. With this instruction, Claude pauses and asks smart questions first. “Who’s the audience?” “What tone are you going for?” “Should this be long-form or a quick summary?”

Those 30 seconds of clarification save you 10 minutes of back-and-forth revisions later. It’s the difference between a one-shot output you can actually use and three rounds of “no, that’s not what I meant.”

“Present a plan before doing the work.”

This is like asking your team member to outline their approach before they start building. Claude will show you its thinking -here’s what I’m going to create, here’s the structure, here’s my approach. You can course-correct before any work is done, rather than reacting to a finished product that missed the mark.

This is especially powerful for bigger tasks like strategy documents, multi-slide presentations, or research projects. A quick plan check takes 10 seconds and can save you from scrapping an entire output and starting over.

“Do not delete or modify anything without my explicit approval.”

Safety net. Since Cowork mode has real access to your files, this instruction makes sure Claude never overwrites, deletes, or modifies anything without asking you first. It’s like telling a new team member “always check with me before changing the shared drive.” Simple, but it gives you peace of mind.

The Compound Effect

Here’s what most people miss: this setup isn’t just about getting better output today. It’s about building a system that gets better over time.

Every time you add a new writing sample to your Me.md, Claude’s understanding of your voice improves. Every strategy doc you save to your Strategy folder becomes context for future projects. Every “don’t do” you add to your list eliminates one more round of editing.

After a few weeks of this, you’ll notice something shift. Your prompts get shorter -sometimes just a sentence or two -because Claude already has all the context it needs. Your outputs need less editing because Claude already knows your voice and preferences. And you start using Claude for things you never thought possible because the quality threshold has gone up so dramatically.

The gap between “meh” Claude output and “this is scary good” output isn’t about prompt engineering wizardry. It’s these four steps. Thirty minutes of setup that pays dividends on every single task you give Claude from here on out.

Try it this week. Set up the folder, write your Me.md, configure the settings, and set your Global Instructions. Then give Claude the same task you’ve given it before -and see the difference for yourself.

Happy Sunday

MJ ✌️