If you've been hearing the buzz about Claude Cowork and wondering where to actually start, this edition is for you.

Cowork is a feature within Claude’s desktop app. 

Whereas the regular chat lives in a browser tab, Cowork lives on your computer.

It can open your folders, read your files, create documents, and run multi-step work on its own. 

In other words, it's an agent: an AI that takes actions like writing files, organizing folders, carrying a task through to a finished document.

As an assistant, it’s incredibly useful. The setup, however, can get a little confusing for someone jumping in for the first time.

A well-organized Cowork project gets smarter every week. A messy one stays a chatbot with file access.

Today's edition is the setup, start to finish (so you can skip the trial and error) 👇

Want to get the most out of ChatGPT?

ChatGPT is a superpower if you know how to use it correctly.

Discover how HubSpot's guide to AI can elevate both your productivity and creativity to get more things done.

Learn to automate tasks, enhance decision-making, and foster innovation with the power of AI.

Projects are just folders

In Claude's web app, a "project" is a workspace you fill with instructions and uploaded files. 

In Cowork, a project is simpler than that: it's a folder on your computer.

There are three ways to start one:

  1. Start from scratch. Make a new folder on your desktop and point Cowork at it.

  2. Use an existing folder. Already have a folder that houses your work? Point Cowork at that. Everything inside becomes context Claude can read.

  3. Import from the Claude web app. If you've spent months building projects in Claude's browser version (the instructions, the knowledge files, all of it) you can bring a project into Cowork with one click. It lands on your computer as a regular folder.

That third option is killer if you’ve already been using Claude a bit. If you've spent months building up a project in the web app, that's accumulated context. It’s transferred in seconds and paired with an AI that can actually do things with it.

The Three-Folder Setup

Inside the project folder, one structure works for nearly everything: Skills, Outputs, Reference.

Each of these categories will get their own subfolder in the project folder. Let’s explore each 👇

1. Skills

This is arguably the most important folder in the project.

A skill is a plain text file that works like a recipe: step-by-step instructions Claude follows every time it does a specific task. 

(If you missed Tuesday’s edition on how/why to create skills, you may want to start here)

In this subfolder, you’ll want to create one skill file per repeating task that you’re trying to perform within Claude Cowork.

A weekly report has its own file. So does a social post, an email campaign, a client deliverable.

2. Reference

This folder houses everything Claude should know that isn't a recipe or a deliverable: brand voice guides, audience notes, past work, research.

The richer this folder, the more informed every output. Think of it as the project's long-term memory.

3. Outputs

Where finished work lands. Organize it by type (reports, content, presentations) so nothing turns into a pile of untitled documents.

The thing to internalize about all three subfolders: Claude is the one reading this structure. 

When you give it a task, it reads the folder. Better organization in means better work out 🤝

Put it on a schedule!

Once a project is organized, scheduled tasks are where Cowork starts working while you don't.

You can tell Claude to run any task on a recurring schedule (daily, weekly, whatever fits).

Just describe the task and the timing in plain English, and it shows up in the project's Scheduled menu.

Scheduled tasks get a lot more interesting once you connect your other tools. 

Cowork links up with apps like Notion, Google Drive, and Slack through connectors: official plug-ins that let Claude read from and write to the tools you already use.

For example: You could schedule a weekly task that gathers relevant industry updates, formats them into snippets, and sends the batch to a teammate in Slack.

A manual research-and-handoff chore, running on its own.

Your first afternoon…

So, to summarize, here’s the whole setup, in five steps:

  1. Pick one area of your work with a task you repeat every week.

  2. Create a Cowork project for it (import from the Claude web app if you've already built one there)

  3. Add the three folders: Skills, Outputs, Reference.

  4. Have Claude write one skill file for that repeating task (the prompt below does this for you).

  5. Run the task, review the output, and tell Claude to fold your feedback into the skill file.

Then repeat step 5 forever, and you’ve got a self-improving, agentic workspace.

Point Cowork at a folder, paste this in, and fill in the bracket 👇

I want to set up this folder as a Cowork project.

The work this project is for:

[Describe it — e.g., "writing my weekly newsletter," "client reports for my agency," "social content for my brand."]

Do three things:

1. Create three subfolders: Skills, Outputs, and Reference.

2. Ask me 3-5 questions about my most common repeating task — how I do it today, what a great result looks like, and what I always end up fixing by hand.

3. Using my answers, write your first skill file in the Skills folder: a step-by-step recipe you'll follow every time I ask for that task.

Going forward, whenever I give you feedback on work in this project, update the skill file so the feedback sticks.

Run your task once, give feedback, and watch the second version come back closer to done.

Your roundup of the latest model releases and updates from the biggest AI labs.

  1. Anthropic doubles Cowork usage limits. Through July 5, Pro, Max, and Team subscribers get twice the usual Cowork usage at no extra charge. It’s Anthropic's nudge to get more people running desktop agents. If you've been waiting for an excuse to try the setup above, the meter runs twice as long for the next month. (The New Stack)

  1. Claude Fable 5 launches. Anthropic released its most capable public model ever — the first in a new "Mythos-class" tier that sits above Opus, previously Anthropic's top model. It leads nearly every benchmark Anthropic tested, with extra safety guardrails on a small slice of sensitive topics. Practical upshot for this edition: the model doing the work inside Cowork just got a major upgrade. (CNBC)

  2. Apple rebuilds Siri on Gemini, and opens the iPhone to Claude. At WWDC, Apple unveiled a Siri rebuilt on a custom Google Gemini model (a deal reported at roughly $1 billion a year), plus a new Extensions feature in iOS 27 that lets you pick a third-party AI — including Claude — as your iPhone's default assistant. Public betas come in July, full launch this fall. A capable AI assistant is about to land on a couple billion devices. (TechCrunch)

Advertise with Build with AI

Get in front of an audience of professionals using AI day-to-day: founders, engineers, operators, and product builders.

Interested in advertising? Respond to this email for rates and details.

Until next time,

William Ryan

Editor-in-Chief @ Build with AI

PS: Follow me on X for daily updates and AI workflows.