
Live Masterclass: Getting Started with Claude Cowork
Live workshop recap: Learn Claude Cowork from scratch - skills as job descriptions, the What-How-Do-Don't prompting framework, building a Weekly Status Report skill, connectors (MCPs), and security best practices. Hosted by Mark Fershteyn with Cynthia Vazquez.
You're drowning. Meetings stack up. Emails multiply. Slack won't stop pinging. And somewhere between the third status update and the fifth "just following up" message, you wonder - is this what I signed up for?
What if one command pulled your entire week together? Every meeting, every deliverable, every follow-up - organized, summarized, and ready to send. In 30 seconds.
That's what we built live in this workshop.
On March 10, 2026, I hosted a live session with Cynthia Vazquez - introducing Claude Cowork to a room full of non-technical professionals. No code. No terminal. Just Cowork, plain English, and one hour to go from "what is this?" to "I built my first AI skill."
This guide breaks down every concept, framework, and demo from that session - plus the actual prompts you can copy and use today.
Watch the Full Workshop
What You'll Learn
Skills & Connectors
How skills work as 'job descriptions' for AI, and how connectors (MCPs) plug Cowork into Google Calendar, Slack, Gmail, and more.
Weekly Status Report
A live-built skill that pulls your calendar, Slack, and email into one executive summary - the exact prompt we built on camera.
What-How-Do-Don't Framework
A 4-step prompting structure that works for every skill you'll ever write. Copy the template and use it forever.
Claude vs Cowork vs Code
Before we get into skills and connectors, you need the mental model. Claude comes in three flavors - and each one gives the AI a different level of access.
- Claude Chat - The sandbox. Ask questions, brainstorm, analyze text. It can't touch your files or connect to your tools.
- Claude Cowork - Your computer. It reads files, creates documents, connects to Google Calendar, Slack, and Gmail. Guardrails are built in. This is what we used in the workshop.
- Claude Code - Full access. Reads files, writes code, runs terminal commands, builds entire applications. This is for builders who want to create software.
Most people reading this guide want Cowork. It's the sweet spot - powerful enough to automate real work, accessible enough that you don't need a CS degree.
Want the full breakdown? Our Claude vs Cowork vs Code comparison guide covers pricing, use cases, and exactly which tool fits your workflow.
Skills = Job Descriptions for AI
Here's the analogy that clicked for everyone in the workshop:
A skill is a job description. Not the vague kind HR posts on LinkedIn. The kind a great manager writes - specific tasks, clear expectations, explicit boundaries.
Skills are markdown files (.md) that live on your computer. When you invoke a skill, Cowork reads the file and follows the instructions. That's it.
A skill file tells Cowork:
- What to do (the task)
- How to do it (the approach, format, constraints)
- What to do specifically (step-by-step instructions)
- What NOT to do (guardrails, off-limits actions)
Think of skills like writing instructions for a really smart intern on their first day. You wouldn't say "do marketing." You'd say "write a LinkedIn post about our product launch, keep it under 200 words, use our brand voice doc for tone, and don't mention competitor names." That's a skill.
Connectors: How Cowork Plugs Into Your Tools
Skills tell Cowork what to do. Connectors tell Cowork where to do it.
Connectors - technically called MCPs (Model Context Protocol) - are integrations that let Cowork access your existing tools:
- Google Calendar → Read your schedule, find meeting conflicts, pull upcoming events
- Slack → Read channels, search messages, get context from conversations
- Gmail → Scan your inbox, find follow-up threads, pull email content
- Google Drive → Access documents, spreadsheets, presentations
- File system → Read and write files on your local computer
You set up connectors once. Then every skill you build can tap into them.
Setup tip: Connect your calendar, email, and Slack first. Those three connectors unlock 80% of the use cases we covered in the workshop. See our Cowork setup guide for step-by-step instructions.
The "What, How, Do, Don't" Framework
This is the prompting framework we taught in the workshop. It works for one-off prompts and for building reusable skills.
Every effective Cowork instruction has four parts:
WHAT
What is the task? Be specific about the outcome you want. 'Create a weekly status report' not 'help me with reports.'
HOW
How should it approach the task? What sources to use, what format to follow, what tone to write in.
DO
Specific actions to take. Step-by-step instructions, files to read, tools to check, output format.
DON'T
Guardrails. What to avoid, what not to include, what's off-limits. This is where most people skip - and where most problems happen.
Here's the framework applied to a real skill:
What-How-Do-Don't Framework - Weekly Status Report
That's it. Four sections. Every skill you build follows this structure.
Live Demo: Building the Weekly Status Report Skill
This was the centerpiece of the workshop. We built a real skill - live, on camera - that pulls together your entire week in one command.
The Skill We Built
What the Output Looks Like
The skill produces a clean, scannable document organized by project:
One command. 30 seconds. Your entire week - organized and ready to send to your manager, share in standup, or paste into your project tracker.
Power move: Combine this with Cowork's /schedule command to auto-generate this every Friday at 4pm. You'll never manually write a status report again.
5 Use Cases from the Workshop
Beyond the status report, we walked through five powerful ways people are using Cowork right now.
1. Watch Me Once
Do a task manually while Cowork observes. It watches your workflow - what you click, what files you open, what steps you take - and creates a skill from the observation.
Best for: Repetitive tasks you do the same way every time. Expense reports, weekly data pulls, formatting documents.
2. Competitive Intel Briefings
Point Cowork at competitor websites and say: "Track these 5 competitors. Every Monday, create a briefing showing what changed on their sites, any new product announcements, pricing changes, and new content published."
Best for: Founders, product managers, marketing teams who need to stay on top of the market without spending hours browsing competitor sites.
3. Zoom Transcript Processing
Drop a Zoom transcript into your workspace. Cowork extracts action items, key decisions, who said what, open questions, and creates a formatted summary ready to share with your team.
Best for: Anyone drowning in back-to-back meetings who never has time to write proper follow-ups.
4. YouTube Content Repurposing
Give Cowork a YouTube video transcript. It creates a blog post, LinkedIn post, tweet thread, newsletter section, and a set of pull quotes - all from one video.
Best for: Content creators, marketers, and founders who want to turn one piece of content into five without the manual rework.
5. Ad Performance Optimization
Feed Cowork your ad performance data (exported CSV from Google Ads, Meta, etc.). It analyzes trends, flags underperforming campaigns, suggests budget reallocation, and drafts recommendations for your next sprint.
Best for: Marketing teams running paid campaigns who want data-driven insights without waiting for an analyst.
Agents vs Skills
This came up as a question during the workshop, and it's worth clarifying because the terminology trips people up.
So the skill is the recipe and the agent is the chef?
Exactly. Here's the distinction:
- A skill is the instruction set. It's a markdown file that describes what to do, how to do it, and what to avoid. It's static - it sits on your computer waiting to be invoked.
- An agent is the execution engine. When you invoke a skill, Cowork becomes the agent that follows those instructions. The agent reads the skill, connects to your tools, executes the steps, and produces the output.
You write skills. Cowork runs them as an agent. The quality of the output depends almost entirely on the quality of the skill - which is why the What-How-Do-Don't framework matters so much.
The takeaway: Don't obsess over "building agents." Focus on writing great skills. A well-written skill with clear instructions, good context, and strong guardrails will produce excellent results every time.
Security & Privacy: The Practical Guide
This was the most-asked-about topic in the workshop. Here's the framework we shared.
The Google Test
Before you paste anything into Cowork, ask yourself:
"Would I be comfortable if this showed up in a Google search?"
If the answer is no, don't paste it. Simple.
What's Actually Safe
- Anthropic's policy: Paid plans (Pro, Teams, Enterprise) do not use your data to train models. Your prompts and files are not used to improve Claude.
- File access is explicit: Cowork only reads folders you grant permission to. It can't silently access your entire computer.
- Skills are local: Your skill files live on your computer. They're markdown files. They don't get uploaded anywhere.
What to Be Careful With
- Credentials and API keys - Never paste passwords, tokens, or secrets into any AI tool.
- Confidential client data - If it's covered by NDA, don't put it in Cowork. Summarize or anonymize instead.
- Personal financial details - Bank accounts, SSNs, credit card numbers are obvious no-gos.
- Private medical or legal information - Keep HIPAA and attorney-client privileged information out.
Cynthia's Approach
Cynthia shared a practical framework during the workshop: use Cowork for process, not secrets. Build skills around workflows, templates, and structures. Feed in anonymized or public data. Keep the sensitive details in your head or in encrypted storage.
Enterprise users: If your company handles sensitive data at scale, look into Anthropic's Enterprise plan. It offers dedicated instances, custom data handling agreements, and SOC 2 compliance. For most individual users, the Pro plan with the Google test is sufficient.
Start Today: Your First Cowork Skill in 5 Minutes
You don't need to build the weekly status report skill. Start simpler. Here's a prompt you can paste into Cowork right now:
That's your first skill. Five minutes. And you'll never start a Monday without knowing exactly what's ahead.
Want more prompts? Our 25 Claude Cowork Prompts guide has copy-paste prompts for file management, research, marketing, operations, and career development.
Need the full setup walkthrough? The Claude Cowork Complete Setup Guide takes you from download to first skill in under 15 minutes.
Want best practices? Our 10 Cowork Best Practices guide covers the patterns that separate great results from mediocre ones.
What will you build? Every skill starts with one task you're tired of doing manually. Pick the thing you do every week that drains your energy. Describe it clearly. Let Cowork handle it. That's how it starts. One skill. One less thing on your plate. Then another. Then another. Before you know it, you've got an AI employee that knows exactly how you work.
Ready to build agents that do the work?
The AI Agents Course: master Claude Code, Claude Cowork, and ChatGPT Codex, and leave with 5+ working agents built on your own work.
- Self-paced + live builds
- 5+ working agents
- 12 months community
Frequently Asked Questions
Is my data safe when I use Claude Cowork?
Anthropic does not train on data from paid plans (Pro, Teams, Enterprise). Your prompts and files are not used to improve Claude's models. That said, always apply the 'Google test' - don't paste anything into Cowork that you wouldn't want appearing in a Google search. Avoid sensitive credentials, private financial data, or confidential client information unless you're on an Enterprise plan with specific data handling agreements.
What's the difference between a skill and an agent?
A skill is the instruction set - a markdown file that describes a task, how to do it, and what to avoid. An agent is the execution engine that runs the skill. Think of it like a recipe (skill) and a chef (agent). The chef follows the recipe, but the recipe is what determines the outcome.
Can I schedule Cowork to run skills automatically?
Yes. Cowork supports scheduling through the /schedule command. You can set skills to run on a recurring schedule - daily morning briefings, weekly status reports, or monthly competitive analyses. The agent runs the skill at the scheduled time and saves the output to your specified folder.
What is Watch Me Once and how does it work?
Watch Me Once is a Cowork feature where you perform a task manually while Cowork observes. It records your workflow - what you clicked, what files you opened, what steps you took - and then creates a skill from that observation. It's the fastest way to automate repetitive tasks because you don't have to write any instructions. Just do the task once and Cowork learns it.
Do I need coding experience to use Claude Cowork?
No. Cowork is designed specifically for non-technical users. Everything in this workshop - from setting up skills to connecting tools - uses plain English. If you can write a clear set of instructions for a human assistant, you can use Cowork.
What are connectors (MCPs) in Claude Cowork?
Connectors are integrations that let Cowork access your existing tools - Google Calendar, Slack, Gmail, Google Drive, and more. MCP stands for Model Context Protocol, but you don't need to understand the technical details. Think of them as bridges between Cowork and the apps you already use.
How is Claude Cowork different from ChatGPT?
ChatGPT is a chat interface - you ask questions and get answers. Cowork is a task executor - it reads your files, follows structured instructions, creates documents, connects to your tools, and runs recurring workflows. It's the difference between talking to an assistant and having one that actually does the work.
What model should I use with Cowork for best results?
Select Opus 4.6 with Extended Thinking enabled. This gives Cowork the strongest reasoning and planning capabilities. For simpler tasks, Sonnet works fine and is faster, but for anything involving multi-step workflows or complex analysis, Opus is worth the wait.
Finished reading?
Take it with you.
Get a downloadable checklist summarizing the key points from Live Masterclass: Getting Started with Claude Cowork, plus weekly AI tips.
Guide Checklist
PDF + Email Series
Related Guides
Claude Cowork Complete Setup Guide: From Install to Automation (2026)
Set up Claude Cowork in 10 minutes. Step-by-step guide covering installation, folder structure, context files, Global Instructions, plugins, connectors, and scheduled tasks.
25 Claude Cowork Prompts That Actually Work (Copy & Paste)
25 ready-to-use Claude Cowork prompts across 5 categories: file management, research, marketing, operations, and career. Copy, paste, and watch Cowork execute.
Claude Cowork vs Claude Code: Which to Use in 2026
Claude Cowork vs Claude Code, settled. Cowork works the files you already have; Code builds things that run on their own. Decision tree, feature table, pricing.
10 Claude Cowork Best Practices (From 100+ Hours of Testing)
10 best practices that separate great Claude Cowork results from mediocre ones. Copy-paste examples for each practice. No fluff, just what works.