Most newsletter creators are flying blind about their audience.
You know what you publish. You track open rates. But do you know exactly what content gaps you’re missing? How your writing style compares to the best newsletters in your niche? What your audience’s deepest frustrations are, written out in specific language you can use in your copy?
A manual competitive analysis could take you days. And it still wouldn’t be as thorough as what I’m about to show you.
In a recent episode of One Shot Show, I sat down with Wyndo from The AI Maker to run a live demo using Claude Cowork, Claude in Chrome, and Firecrawl MCP.
The Claude Code Workshop goal: generate a complete newsletter intelligence report on genaiunplugged.substack.com without writing a single line of code or opening a terminal.
The result was a 17-page Word document covering audience profile, writing style DNA, content gaps, growth strategy, and 10 specific content ideas.
If you’ve been intimidated by terminal-based AI tools, this workflow was built for you. By the end of this article, you’ll have the exact setup, the prompt structure, and a clear picture of what the output looks like.
You don’t need a terminal, a developer, or three days of setup to generate competitive newsletter intelligence. You need a goal, a URL, and 15 minutes.
“For chatbot, you still need to have a very long prompt. And you still need to become the middleman where you need to copy-paste everything. For agents, you just describe your goal and intent, set up some constraints or guardrails, and let the agent do the work.”. Wyndo
👋 Julley, I'm Dheeraj, an AI systems builder.
I build production-grade AI systems at work by day and ship my own products by night, 9 and counting, including SubflowAI and the Content OS Agents Toolkit. This newsletter is the bridge between those two worlds. Every system, every build, documented step by step.
Join 1,800+ builders getting the exact AI setups, prompts, and workflows that actually work in your business.
What Is Claude Cowork and Why Does It Matter Now?
Claude Cowork is Claude Code’s agentic engine wrapped in a graphical user interface inside Claude Desktop. You get the same powerful, multi-step automation that developers use in the terminal, but through a point-and-click interface that requires zero technical background.
As Wyndo explains at 4:01, we’re now in the third era of AI tools:
Chatbots (Claude.ai, ChatGPT): You ask questions, get text back, copy-paste the output yourself
Claude Code: Agentic, multi-step automation, but lives in the terminal
Claude Cowork: Agentic automation with a GUI, built for everyone else
The distinction matters. Chatbots make you the middleman. You copy the output, paste it somewhere, run another prompt, copy again. Agents take a goal and execute it end-to-end. Cowork brings that capability to people who have never opened a terminal.
Both Wyndo and I are self-described non-coders who use Claude Code regularly. Cowork lowers the bar even further. And for newsletter creators who want competitive intelligence without a technical setup, that bar matters.
GenAI Unplugged is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.
What Are Claude Cowork’s 5 Core Capabilities?
Claude Cowork has five capabilities that separate it from a standard chatbot, and each one plays a role in the newsletter analysis workflow. Wyndo walks through all five starting at 7:46.
Claude.md (the project brain)
A markdown file you create inside your project folder. It stores your audience profile, subscriber count, the problem you solve, your content themes, and any other context the agent should know before starting work. Think of Claude.md as the briefing document your agent reads before every task.
Planning and to-do lists
When you submit a task, Cowork doesn’t start executing immediately. It creates a visible checklist first, shows you what it’s planning to do, and checks items off as it goes. You can see exactly what the agent is doing at every step.
Claude in Chrome
A browser extension that gives Cowork visual access to the web. The agent takes screenshots of web pages, reads what it sees, and navigates autonomously. It’s not reading HTML source code. It’s looking at the page the way a human would.
Parallel sub-agents
When you use Opus 4.6, Cowork automatically deploys multiple sub-agents to work on different parts of a task simultaneously. Sonnet 4.6 can work too, but requires you to click the model selector and toggle Extended Thinking ON.
File outputs
Cowork can generate real files directly: Word documents, Excel spreadsheets, PowerPoint presentations, HTML. No plugins required.
Commercial break: Claude Code Builder cohort
The founding batch of my Claude Code cohort starts in July on Maven. Four weeks, six sessions. Learn to build the AI system with me.
Only 12 Seats. When they’re gone, the founding price ($797) closes and Cohort 2 opens at $1,597.
Use code GENAI20 for 20% off. Expires July 9. Check the Syllabus →
How Do You Set Up Claude Cowork for Newsletter Analysis?
Three tools need to be configured before you run the newsletter DNA extraction. Wyndo covers the setup at 22:00.
What you need:
Claude Cowork (inside Claude Desktop)
Download Claude Desktop from claude.ai/download
Open Claude Desktop > Settings (gear icon) > Features > toggle “Cowork” ON
Create a new project folder, then click “Open in Cowork” from the project menu
Setup time: 10-15 minutes
Claude in Chrome
Install from Chrome Web Store (search “Claude in Chrome”)
Open the extension and sign in with your Anthropic account
Must be installed AND logged in before Cowork can trigger it
Setup time: 2 minutes
Firecrawl MCP (Model Context Protocol)
Sign up at firecrawl.dev and copy your API key
In the Cowork sidebar, click Connectors > Add > search “Firecrawl” > paste your API key
Setup time: 5 minutes
Note: Firecrawl changed from 500 credits/month to 500 credits total. Plan for this.
Free fallback tools to configure:
Jina.ai (free scraping, add as fallback in your prompt)
Claude’s built-in web fetch (a text-only scraping capability that pulls page content but can’t see visual layouts - automatic fallback if both tools fail)
One important note: Cowork is restricted to your home directory. Claude Code has full system-wide folder access. If you need to reach files outside your home folder, Claude Code is the better choice for that task.
How Do You Build the Newsletter DNA Extraction Prompt?
The prompt structure is the key piece. At 20:01, Wyndo reveals what goes into it and why each element matters.
You are a newsletter analyst. You have access to:
- Claude in Chrome (for visual browsing)
- Firecrawl MCP (for full content scraping)
- If Firecrawl fails, use Jina.ai. If Jina fails, use Claude built-in web fetch.
Analyze this newsletter: [NEWSLETTER URL]
Phase 1: Homepage
Use Claude in Chrome to screenshot and analyze the homepage.
Extract: newsletter name, description, audience signals, post count.
Phase 2: Archive
Navigate to [URL]/archive?sort=top
Identify the top 10 posts by engagement.
DO NOT access paywalled posts. skip them and move to the next.
Phase 3: Content Scraping
Use Firecrawl to scrape the full text of 5-7 top free posts.
For each post extract: title, full body, topics covered, writing style.
Phase 4: Analysis
Synthesize across all posts to identify:
- Audience profile (who they are, frustrations, aspirations)
- Problem solved
- Content themes and frequency
- Writing style and voice characteristics
- Content gaps and opportunities
- 10 specific content ideas based on patterns
- SEO and niche positioning
- Competitive positioning
- Top 5 immediate actions
Phase 5: Output
Generate a complete Word document with all findings.
Include an executive summary on page 1.A few notes on this structure. The prompt explicitly names which tools are available. The agent won’t assume. You must declare them.
The fallback chain is built directly into the prompt so the agent knows exactly what to do if one tool fails. Paywalled content is excluded on purpose: partial scraped content creates messy, incomplete analysis.
Wyndo recommends using Opus 4.6 with extended thinking ON. This is a token-heavy workflow. On Claude Max plan, that’s managed. On API billing, costs accumulate faster than you’d expect.
What Happens When the Agent Runs?
At 33:43, the live demo shows what execution looks like. The agent doesn’t start scraping immediately. It creates a planning checklist first. You see items like “Visit homepage,” “Navigate to archive,” “Identify top posts,” “Scrape post 1 of 7.”
Then Claude in Chrome takes over the browser. Wyndo and I both agreed: watching this in real time is genuinely impressive. The agent opens a browser tab, screenshots the page, reads it, clicks through to posts, and navigates the archive without any input from you.
Midway through the live demo, Firecrawl ran out of credits. The agent switched automatically to Claude’s built-in web fetch and continued without stopping.
“Whatever tools that you use, make sure you have a system of fallback whenever certain tools don’t work. It needs to reset or shift to another tool that is also reliable for you.”. Wyndo (at 37:46)
One quirk: Cowork has a UI display bug where the progress indicator stops updating but execution continues in the background. If it looks frozen, give it time. The work is still happening.
What’s Inside the 17-Page Newsletter Report
At 43:29, Wyndo walks through the full report generated for genaiunplugged.substack.com. The 17-page Word document included:
Audience Intelligence
Detailed audience profile: who reads the newsletter, their frustrations, what they already know, what they want to learn
Problem solved list: specific pain points the newsletter addresses
Audience segmentation signals
Content Analysis
Recurring themes across top posts with frequency data
Structure patterns in high-performing posts
Writing style and voice characteristics
Vocabulary and tone analysis
Gap and Opportunity Mapping
Content gaps: topics the audience wants but the newsletter hasn’t covered
10 specific content ideas generated from analyzing existing post patterns
These aren’t generic suggestions. They’re derived from what’s working.
Growth Strategy
Newsletter SEO analysis and keyword territory
Niche clarity assessment
Competitive positioning recommendations
Top 5 immediate actions
Bonus: The agent independently identified SubflowAI from analyzing the posts, without being told about it. The competitive intelligence section included product positioning analysis for a product Wyndo never mentioned in the prompt.
As Wyndo notes, the audience profile section alone is worth the workflow. You can feed those insights directly back into your newsletter’s About page, welcome email, and Claude.md file.
Should Claude Code Users Switch to Claude Cowork?
The honest answer, discussed at 48:08 by both of us, is no. Not if you’re already a Claude Code user.
Claude Code is faster. It has full system-wide folder access. Your workflows are already established there. Switching to Cowork for a GUI adds friction, not value.
Where Cowork makes sense for Claude Code users
One-off tasks you don’t want to set up a full Code project for
Tasks where someone else will run the workflow (a non-technical team member or client)
Where Cowork is the clear winner
Non-technical users who find terminals daunting
Newsletter creators who want agentic AI without a learning curve
Anyone who wants the output files handled automatically
On the question of mobile access of Cowork Dispatch:
Cowork has a Dispatch feature for triggering agents from your phone while your computer is awake. For Claude Code users, Claude Code Channels (Telegram and Discord integration) achieves similar always-on access with more flexibility. We’ll cover that comparison in Episode 5 of One Shot Show.
“In Claude Cowork, somebody who comes in as a non-technical background can completely get similar capabilities and do the jobs, without any code being shown.”. Wyndo (at 51:31)
Common Questions About Claude Cowork Setup
What is Claude Cowork and how is it different from Claude AI on the website?
Claude Cowork is built on Claude Code’s agentic engine with a graphical interface inside Claude Desktop. Unlike Claude.ai, Cowork can create plans, access your computer’s files, control your browser via Claude in Chrome, deploy parallel sub-agents, and generate real files (Word, Excel, PowerPoint). The chatbot gives you text back; Cowork executes multi-step tasks end-to-end.
Do I need to know how to code or use a terminal to use Claude Cowork?
No. Cowork was designed for non-technical users. You interact through a graphical interface, never see a terminal, and describe your goals in plain language. Both Wyndo and I are self-described non-coders who use Claude Code and Cowork successfully.
What tools do I need before running newsletter analysis with Claude Cowork?
Three tools: Claude Desktop with Cowork enabled, the Claude in Chrome browser extension (installed and logged in), and Firecrawl MCP configured via the Connectors panel. Sign up at firecrawl.dev, get your API key, and add it as a connector. Jina.ai is a free fallback if Firecrawl credits run out.
What model should I use for newsletter DNA extraction?
Opus 4.6 with extended thinking ON is recommended. It automatically deploys parallel sub-agents for complex tasks and produces the most comprehensive output. Sonnet 4.6 can work but requires manually toggling thinking mode. This is a token-heavy workflow, so Claude Max plan users have an advantage over API billing.
Can I analyze a competitor’s Substack, not just my own?
Yes. The workflow is URL-based. Swap in any competitor’s public Substack URL. Paywalled content will be skipped, but the analysis of free posts and archive patterns still generates a useful competitive profile.
What happens if Firecrawl runs out of credits mid-workflow?
Claude’s built-in web fetch serves as an automatic fallback. This happened live during our demo and the workflow continued without interruption. For more reliable coverage, configure Jina.ai (free) as the first fallback in your prompt instructions.
How does Cowork handle memory between sessions?
Cowork doesn’t have native session memory. Instead, create a memory.md file inside your project folder and instruct the agent to log progress there daily. The agent reads this file before new tasks. Global instructions in Cowork settings also maintain project context across sessions.
Key Takeaways
Claude Cowork is Claude Code with a GUI: Same agentic engine, no terminal required, built for non-technical users who want multi-step automation.
Claude.md is the foundation: Set it up before anything else. Keep it under 150 lines. Reference external files instead of pasting full content (progressive disclosure).
Always build a fallback chain: Firecrawl → Jina.ai → Claude built-in web fetch. State the fallback order directly in your prompt so the agent knows what to do when one tool fails.
Target /archive?sort=top, not the homepage: The archive sorted by top posts gives you the most accurate signal on what content performs.
Skip paywalled content in your prompt: Partial paywall scrapes create noisy, incomplete analysis. Explicitly tell the agent to skip and move on.
The output is immediately actionable: The audience profile section feeds directly into your About page, welcome email, and Claude.md file.
Claude Code users don’t need to switch: Cowork is best for non-technical users and one-off tasks. Power users should stay in Code.
Resources Mentioned
Claude Cowork: Claude Code’s agentic engine with GUI, inside Claude Desktop. Released January 2025.
Claude in Chrome: Browser extension required for Cowork to take screenshots and navigate the web autonomously.
Firecrawl MCP: Web scraping connector for Cowork. Sign up at firecrawl.dev. 500 one-time credits.
Jina.ai: Free web scraping alternative to Firecrawl. Use as fallback.
Claude Desktop: The application that hosts Claude Cowork.
Opus 4.6: Recommended model for newsletter DNA extraction. Deploys parallel sub-agents automatically.
Sonnet 4.6: Lighter alternative. Requires manual thinking mode toggle.
Claude Code: Terminal-based agentic AI. The underlying engine of Cowork. Still preferred by power users.
Claude Code Channels: Telegram and Discord integration for Claude Code. Coming in Episode 5 of One Shot Show.
Gemini CLI (Flash 2.5): Free search and web fetch fallback I use in my Claude Code setup.
SubflowAI: My Substack Notes scheduling product, which the agent independently identified from post analysis.
One Shot Show: The live show Wyndo and I co-host on Substack. Episode 4 covers Claude Code vs n8n. Episode 5 covers Claude Code Channels vs OpenClaw.
Your 15-Minute Challenge
Open Claude Desktop and set up Cowork if you haven’t. Install Claude in Chrome and confirm you’re logged in. Then write a single Claude.md file for one newsletter project you’re working on. Include your audience profile, the problem you solve, and three content themes. Keep it under 150 lines.
That’s your foundation. A Claude.md file done well is the difference between an agent that produces generic output and one that generates insights specific to your newsletter.
Once your Claude.md is ready, paste the prompt from the previous section into Cowork, swap in your newsletter URL, and run it. You’ll have your 17-page report within the hour.
Watch the full One Shot Show episode for the live demo, the exact prompt structure, and the full 17-page output walkthrough.
Next up: how to trigger Claude Code agents from your phone via Telegram and Discord. No Cowork session limitations, no desktop required. That’s Episode 5 of One Shot Show.
