The Claude Cowork Setup That Saves Hours

Most AI sessions feel like Groundhog Day. You explain who you are, what your business does, your tone, your goals, your no-go words. Every. Single. Time. By the time the model finally gets you, twenty minutes are gone and you haven’t done any real work.

Then I came across this brilliant Cowork setup from an AI professional on LinkedIn, and it solved the whole context problem in one shot. The creator laid out an 11-step blueprint that gets Claude to already know you before you type your first prompt. Once it’s in place, you write a single line and the model understands your voice, your business, and your style preferences.

I’ve tried versions of this before, but the structure this expert shared is the cleanest one I’ve seen. Here’s the full breakdown with the why behind each move.

The 11-step Cowork setup, with rationale

1. Grab the Claude desktop app, not the browser version. The desktop app is the only place Cowork lives. The browser tab gives you chat. The app gives you file system access, folders, and the global instructions panel you’ll need later.
2. Pay $20 a month for Pro. Skip the $100 plan. The original poster is direct about this: the cheaper tier covers everything in this workflow. The bigger plan is for heavy team usage, not for getting Cowork running.
3. Create one folder called “Claude Cowork”. One root folder keeps the whole system contained. No scattered files, no “where did I save that prompt” moments. Everything you do with Claude lives under this one roof.
4. Inside it, make three subfolders: About me, Outputs, Templates. The author’s logic here is clean. About me holds your context. Outputs holds anything Claude generates so you can reference past work. Templates holds reusable prompt structures you don’t want to retype.
5. Inside About me, create three .md files: about-me, my-company, anti-ai-writing-style. This is the part most people skip and then wonder why their AI output sounds generic. Splitting personal context from company context from voice rules means Claude can pull the right one for the right task. Writing a personal note? It uses about-me. Drafting a sales page? It pulls my-company plus the anti-AI style file.
6. Let Claude interview you using AskUserQuestion. Staring at a blank .md file is the fastest way to write a boring profile. The expert recommends letting Claude ask you questions one at a time. You answer, it drafts, you refine. The output is sharper than anything you’d write cold.
7. Dictate your answers with Wispr.ai for speed. Talking is roughly three times faster than typing, and it captures the way you actually sound. The original poster flags this as a small upgrade with outsized impact, and I agree. Your voice file should sound like you, not like a corporate bio.
8. Keep each file under 2,000 words. Long files dilute focus. Short, dense files give Claude exactly what it needs without burning context budget on filler. If you can’t say it in 2,000 words, you probably haven’t decided what’s important yet.
9. Open Settings, find Cowork, then Global Instructions. Paste this line: “Always read files in ‘About Me’ before tasks”. This is the lock that holds the whole system together. Without it, Claude sees the files but doesn’t always pull from them. With it, every task starts by loading your context. The creator calls this the non-negotiable step, and it absolutely is.
10. Pick Opus 4.7 with Extended Thinking turned on. Opus handles nuance better than the lighter models, and Extended Thinking gives it the reasoning runway to actually use the context you just loaded. For one-shot quick answers, lighter models are fine. For real work, this combo is the move.
11. Write a one-line prompt. The model already knows you. Instead of “You are a marketing expert helping me write a LinkedIn post in my voice about X for an audience of Y”, you type “Write me a LinkedIn post about X.” Claude pulls the voice, the audience, the company context, the style rules, all on its own.

Why this setup is worth a Saturday afternoon

The whole reason most people give up on AI is the friction of restarting context every session. Kill the friction once, save it forever.

I was honestly surprised at how big a difference Step 9 makes. Without that single line in Global Instructions, the rest of the structure is just neat folders. With it, every chat starts pre-loaded. The author of the post nailed the sequencing, and that one detail is the difference between a system that works and a system that sits in a folder unused.

The other underrated piece is the anti-ai-writing-style.md file. Most setups skip the negative rules. Spelling out what you don’t want, the words you ban, the phrases that make you cringe, the sentence structures you avoid, gives Claude a much sharper edge than just listing positive style traits. This savvy professional clearly learned that the hard way and shared the shortcut.

Quick tips before you start

• Write the about-me file first, then my-company, then style. Voice rules need real context to work against.
• Re-read your .md files every couple of months. Your business changes. Your voice changes. The files should too.
• If a Claude output feels off, the fix is almost always in your About me files, not in the prompt.
• Keep the Outputs folder organized by date or project from day one. It compounds fast.

Check the original LinkedIn post for the full setup straight from the source.