How I Set Up Claude Cowork (So It Actually Works For Me)

Viktor is an AI coworker that lives in Slack, right where your team already works.

Message Viktor like a teammate: "pull last quarter's revenue by channel," or "build a dashboard for our board meeting."

Viktor connects to your tools, does the work, and delivers the actual report, spreadsheet, or dashboard. Not a summary. The real thing.

There’s no new software to adopt and no one to train.

Most teams start with one task. Within a week, Viktor is handling half of their ops.

Add Viktor to Slack for free.

At the end of January, I wrote about setting up Moltbot (now OpenClaw), an open-source AI agent that promised persistent memory and proactive check-ins.

I spent three days on terminal commands, VPS setup, and Slack integrations. I got it partially working, but it required more technical chops than I wanted to develop, and running it on a remote server limited what I could actually connect it to without those chops.

I liked the idea of an AI agent that could do things rather than just generate text.

So when Anthropic released Claude Cowork for Windows in February 2026, I was ready.

It was less technical, it ran on my actual computer, and it could work with my local files.

And the first thing I realized? It didn't remember me.

I opened Cowork and it was just... Claude.

Except worse, because it started from zero every time. No idea who I was, what I was working on, or how I liked things done.

Then I found an article by Hannah Stulberg about using Claude Code as a personal assistant. The key insight was simple: you can teach Claude who you are by creating a CLAUDE.md file. It's a plain text file that Claude reads at the start of every session.

Think of it like handing a new employee an onboarding doc on their first day, except this employee reads it perfectly every single time.

I realized the same principle would work for Cowork. So instead of just jumping in and asking it to do stuff, I spent time building the foundation.

Here's what that looked like:

Step 1: Create a dedicated folder on your computer.

Cowork needs a local folder to work in. If you're used to doing everything in Google Drive like I was, this feels a little annoying at first. But it's the best way to keep things contained and limit risk.

I created a new folder in my Documents, then set it to get backed up by Google Drive so I'd have a copy if something went sideways.

Step 2: Build a folder structure that makes sense for your work.

The cool thing about doing this with Cowork is that I could have a chat with it about the best folder structure, and then it built the structure. It created folders and moved them around when needed.

I organized mine around the different areas of my life and business: business/life planning, content, consulting clients, inputs (where raw transcripts and data land), outputs (where finished deliverables go), and a system folder for memory files and scripts.

Your structure will look different, but the principle is the same: give Claude a map of your world so it knows where things live.

Step 3: Write your CLAUDE.md file.

This is the single most important thing you can do. It's a text file at the root of your workspace that tells Claude who you are, what you're working on, how you like things done, and where to find stuff.

Mine includes my name, location, what I do, my writing style preferences, file naming conventions, and rules like "ask before sending anything external." Think of it as the operating manual for your AI partner.

You don't have to write this from scratch. I literally told Claude what I wanted it to know about me and asked it to create the file.

It built the first version, I reviewed it, told it what to change, and within a few minutes I had a solid CLAUDE.md that it reads every time it starts up.

Step 4: Add CLAUDE.md files to each folder.

This is the move that made everything click. Each project folder gets its own instruction file explaining what's in there and how Claude should work with it.

My newsletter folder has one that describes the audience, the structure, the voice, and the workflow. My client folder has contacts and project status. When Claude opens a project, it reads the local instructions and immediately has context.

Same trick here: once I had the root CLAUDE.md working, I just asked Claude to create one for each folder based on what was in there.

Since I already had Claude Projects for most of these work areas, I copied the project instructions and gave them to Claude Cowork so I didn’t have to repeat myself.

It reviewed the files and project instructions, determined each folder’s purpose, and wrote the instructions.

Step 5: Loaded my files

This was the boring part: downloading the files I wanted Cowork to access.

My first focus was finally bringing all my content into one place so I could index it and have Claude help me repurpose it, find connections, and come up with ideas for new content.

Some files I just had to download manually. But I had dozens of coaching calls in Google Docs.

Luckily, there’s a connector for that. I just had to tell Cowork where the folder was, give it clear instructions not to delete anything, and it downloaded all those files as text files and saved them in the right folder for me.

A hundred plus emails I’d sent: same thing. I connected Cowork to my Gmail account, told it which label to look for, and it downloaded all of them while I did other things.

Lastly, I had Cowork rename everything with my naming convention (every file I create starts with the date - YYYY.MM.DD).

Step 6: Set up a memory system.

Every session gets a daily memory entry: what we worked on, files created or modified, key decisions, and context for next time.

When I start a new session, Claude reads yesterday's notes and picks up where we left off. No "remind me what we were doing" conversations. No lost context. It just knows.

And when I want to go back and piece together how I did something (for example, for this newsletter), Claude can read the daily memory and tell me exactly when I did what.

Most AI content still focuses on prompts: What to type, how to phrase it, which model to use.

That stuff matters, but it's not what separates people who are "using AI" from people who are becoming AI operators.

And if you want to become an AI operator, the foundation matters.

The folder structure, the instruction files, the memory system. The boring infrastructure that makes every future session dramatically more productive than the last.

I set this up over the course of a couple days. It wasn’t flashy or exciting. Just a few hours spent organizing files, writing instructions, and teaching an AI how I work.

But here's what happened after that: every session since has started with full context. Claude knows my projects, my clients, my writing style, my preferences.

It reads yesterday's memory and today's priorities. And because each session builds on the last, the value compounds over time.

Week 1 with Cowork was fine. Week 6 was dramatically better, not because the tool improved, but because I'd invested in the foundation.

If you're thinking about setting up an AI agent, don't start with the cool stuff. Start with the boring stuff. Build the folder. Write the instructions. Create the memory system.

Then watch what happens.

Thanks for reading!

Nathan Rodgers

👋 Say hello on Substack and LinkedIn

Were you forwarded this email? Click here to subscribe.