Claude Cowork Complete Course

The engineers already have Claude Code.

It changed how they build. Completely. They don't write code the same way anymore. They don't think about software the same way anymore.

And for months, the rest of us, the marketers, the founders, the ops people, the sales teams, watched from the sidelines.

That changes with Claude Cowork.

Claude Cowork is Claude Code for people who don't code. And if you're reading this thinking "I've opened Cowork before, it looked like a chat," you're not using it. Not really.

This is the guide I wish I had on day one. Every feature. Every setup step. Every first prompt.

Here's what we're going to cover:

I also did a full breakdown video that complements well this guide:

Now let's start.

The regular Claude chat is a question-and-answer machine. You ask. It responds. That's the whole thing.

Cowork is an AI agent. It takes action. It reads your actual files. It writes back to them. It connects to your tools. It can spin up other AI instances to do work in parallel while you watch.

Think of it this way:

The interface looks almost identical. Under the hood, it's a completely different tool.

(Claude Code, if you've heard of it, works similarly but is designed more for engineering tasks and it’s overall harder to understand without a technical background).

Start like this every time:

The key move is forcing Cowork to ask you questions before it does anything. It generates a form. You click answers. It executes with context instead of guessing.

This one change produces better outputs than any prompt hack I've ever tried.

Every bad AI output has the same root cause: not enough context.

Cowork solves this differently from every other AI tool. When you select a folder, Claude gets direct access to every file inside it. It can read them, create new files, edit existing ones, and save outputs, all inside that same folder.

Here's why this matters:

In regular Claude chat, every conversation starts from zero. You have to re-explain who you are, what you're working on, and what good output looks like.

Every. Single. Time.

In Cowork with a proper folder setup, you open a chat and Claude already knows everything. It read your files before you typed a word.

Start with these (I’ll show you how to create them a few lines below). You can build more over time.

You don't need all of these on day one. Start with one or two. The system improves as you add more.

You don't have to write them from scratch. Ask Claude to do it:

One of these files, done properly, is worth more than a hundred prompts.

If you have multiple types of work, marketing tasks, sales tasks, client work, Projects let you keep them separated with their own context.

A Project is a saved workspace with:

How to set one up:

Example:

Now every time you open that Project, Claude already knows the context. You just talk.

Connectors let Cowork take actions inside your other tools. Not just read about them. Act inside them.

Connected to Gmail? Claude can check your inbox, draft replies, and send emails. Connected to Notion? It can create pages, update databases, and pull in content. Connected to your CRM? It can research leads, log notes, and update deal stages.

One tip: go to each connector's settings and set it to "Always Allow." Otherwise Claude will ask for permission every time it uses it. Gets annoying fast.

Most tools that aren't natively in Cowork have an MCP server available.

MCP (Model Context Protocol) is just a package that gives Claude structured access to a tool's features. All the native connectors are MCP servers under the hood,  they're just pre-installed for you.

For tools not on the list:

If your tool has no MCP server at all, ask Claude to build one:

Cowork can also open a browser and interact with websites. Or, in extreme cases, take control of your entire computer.

Here's my honest take: use these as a last resort, not a first move.

Browser use is token-heavy, slow, and error-prone. Computer use is even more so.

The order of preference:

One exception: if you need to scrape data from social media sites that block Claude's direct access (LinkedIn, Instagram, Facebook), use a tool called Apify. It's a connector with scrapers for every platform Claude can't reach. Install it in Connectors. If you do any kind of lead research or content monitoring, you'll use it constantly.

One more thing, you can trigger Cowork from your phone.

Install the Claude app on your phone. Connect to the same account. Enable Dispatch in settings. Now you can send Cowork tasks while you're away from your computer and the agents work on your desktop in the background.

Sent from my phone. Handled on my computer. By the time I'm back at my desk, it's done.

Skills are folders of instructions, context files, and resources that your AI agent can use to complete a specific process, more accurately, more efficiently, and more consistently than a one-off prompt ever could.

At the core is a skill.md file. Think of it as an SOP. A step-by-step instruction sheet for how to do a specific process.

Around that core, you can add:

The reason you can give one agent access to thousands of skills without it getting confused: a feature called progressive disclosure. Only the skill's name and description live in memory at all times. The full skill only loads when it's triggered. Reference files only load when the skill calls for them.

One agent. Unlimited skills. No context overload.

Step 1: Think before you prompt.

Most people skip this. It's the biggest mistake. Before you build anything, write down:

Spend 10 minutes here. It will save you hours of iteration later.

Step 2: Gather your reference files.

For content and copywriting skills, the files that make the biggest difference are:

If you don't have these yet, create them now with Claude using the prompt from Section 2.

Step 3: Build the skill using this framework:

Step 4: Test it, then improve it.

Skills are never finished. The more you use them, the better they get.

After building, ask Claude:

It spins up sub-agents to run the tests in parallel, then reports back with scores and suggestions. You tell it what to fix. It updates the skill. You run the test again.

A couple of rounds of this and you have a skill that's genuinely good.

Just to give you a example, I created this Humanizer Skill that removes that typical AI-generated writing patterns from text.

Plugins are bundled sets of skills, connectors, and commands packaged together into a specialist.

Install the Sales plugin and Cowork becomes a sales expert, it knows how to do account research, call prep, outreach drafting, and pipeline review.

Install the Marketing plugin and it knows content creation, campaign planning, competitor analysis.

Anthropic has built over 10 official plugins across the most common business functions: Sales, Marketing, Legal, Finance, Data, Product, Customer Support, and more.

Anthropic's plugins are built to be general. They don't know your process, your ICP, or your tone of voice.

Building your own is the real unlock.

The approach I recommend:

To build a plugin:

Claude builds it. Then you can either install it directly or ask for a zip file to share with your team.

One person's expertise, built into a skill, can now be used by everyone in the company. Your best salesperson's outreach process becomes a skill. Your best marketer's content framework becomes a plugin. Every new hire gets access to that on day one.

Scheduled tasks let you set a prompt or skill to run automatically, every hour, every day, every week, at a specific time.

When your desktop app is open, the task fires. Cowork executes the skill, does the work, and the results are waiting for you.

Here are the ones I run:

Every one of these used to require my attention. Now they happen while I'm elsewhere.

Option 1: From the sidebar:

Option 2: From a chat: Type /schedule in any chat and describe the task.

Option 3: When creating a skill: Tell Claude to schedule it at the same time. Example:

It builds the skill and adds it to the scheduled section automatically.

Scheduled tasks only run when your Claude desktop app is open. If the app is closed when the task should fire, it runs immediately when you next open it.

For truly autonomous, always-on automations, especially non-human-in-the-loop ones, a platform like n8n or Make is still the better option. Cowork's scheduling is best for workflows where you want to be available to review or guide the output.

Open your calendar. Book this time now. Come back to this guide when you're ready.

Minutes 0-5: Install.

→ Go to claude.com/download. Download the desktop app. 

→ Sign in. Get a Pro account ($20/month). 

→ Open the app. Click the Cowork tab.

Minutes 5-15: Build your first context file.

→ Open a text editor. Create a new file called about-me.md. 

→ Write three things: what you do, who you serve, one example of work you're proud of. 

→ Save it somewhere accessible.

→ Now open a new Cowork session. Select the folder that file is in. 

→ Paste this prompt:

→ Answer its questions. Let it build your voice file. Now you have two context documents.

Minutes 15-25: Do one real task.

→ Pick something you actually need to do this week. 

→ Paste this prompt:

→ Answer its questions. See what it produces.

Minutes 25-30: Build your first skill.

→ If the output was good, say:

→ Done. Your first skill exists. Come back to it tomorrow and make it better.

The value of this setup isn't in the features.

It's in the context that builds over time.

Every skill you refine, every reference file you add, every decision you log, it all compounds. The AI agent you and your team have after 6 months of using this is fundamentally more capable than the one you start with on day one.

And if your competitor starts 6 months after you, they're not just behind on a tool. They're behind on 6 months of accumulated intelligence that makes the tool actually work well.

The context is the moat. Start building it.

If this guide helped you, be that person for someone who's still starting from scratch every chat. Share it with them.

See you in the next one :)