Claude Code for product marketing [Series 5 of 6]

Win/loss dashboards. Landing page prototypes. Competitive monitoring that runs itself. A searchable objection library for sales.

All things that normally require a developer. All things you can build yourself in minutes – by typing sentences. No code. No engineering ticket.

That’s Claude Code. The name sounds technical. The experience isn’t. You describe what you want in plain English, same as Cowork. The difference is what Claude can build.

What Claude Code actually is

You already know Cowork. You open the Claude desktop app, select a folder, describe what you want, and Claude produces files – a .docx, a .pptx, an analysis saved to your folder.

Claude Code does the same thing, but in a terminal. A terminal is the plain text window on your computer where you can type commands directly to your machine – on a Mac it’s literally called “Terminal,” on Windows it’s called “Command Prompt” or “PowerShell.” It looks like a blank screen with a blinking cursor. No buttons, no menus, no drag and drop. Just text.

That sounds like a downgrade. It’s actually an upgrade – because the terminal gives Claude more power than it has inside the desktop app.

In Cowork, Claude can produce documents.

In Code, Claude can build things: interactive dashboards that open in your browser, automated systems that run on a schedule, tools that pull data from multiple sources and visualize patterns.

It can install software, connect to external tools and databases, and create applications. All from the same interaction model – you type what you want, Claude does it.

A few examples of the difference:

• Cowork produces a competitive battle card as a Word doc. Code builds a competitive tracking dashboard that updates itself every week.
• Cowork synthesizes interview transcripts into a research summary. Code builds an interactive tool your whole team uses to explore the data by theme, segment, and quote.
• Cowork creates a launch brief. Code builds a launch readiness tracker that you update daily by telling Claude what’s changed.

Cowork produces deliverables. Code produces infrastructure.

Everything you’ve built so far carries over. Your Skills work in Code. Your context files apply. Your connectors are available. You’re stepping up to a more powerful surface with the same foundation underneath.

Setting it up

The setup is about 10 minutes.

1. Install Node.js

Go to nodejs.org. Click the download button. Run the installer. Click through the prompts. Done. Node.js is a piece of software that Claude Code needs in order to run – think of it as the engine under the hood. You install it once and never think about it again.

2. Install Claude Code

Open your terminal. On Mac: hit Command+Space, type “Terminal,” and open the app that comes up. On Windows, you’ll need to install WSL (Windows Subsystem for Linux) first – Google “install WSL” and follow Microsoft’s guide. It’s a one-time setup that takes a few minutes.

Once your terminal is open, you’ll see a blank screen with a cursor. Paste this line in and hit Enter:

npm install -g @anthropic-ai/claude-code

Don’t worry about what this means – it’s just the install command. It downloads for about a minute.

FYI: If you see a wall of red “npm error” text mentioning “permission denied” or “EACCES” – don’t panic. This is the most common speed bump. Run this version instead:

sudo npm install -g @anthropic-ai/claude-code

It’ll ask for your Mac password. When you type it, nothing will appear on screen – no dots, no asterisks, nothing. That’s normal. It’s hidden for security. Type your password and hit Enter.

Once you see “added packages” with a cursor back at the bottom, you’re set.

When it finishes, type claude --version and hit Enter. If a version number comes back, you’re set.

3. Point it at your working folder

Claude Code works from whatever folder you launch it in. If you already have a Cowork working folder set up from Part 2, use that. Same folder structure, same context files, same Skills.

On Mac: right-click your folder, pick “New Terminal at Folder.” On Windows: right-click, pick “Open in Terminal.”

Type claude and hit Enter.

It should look something like this:

4. Sign in

Just use the arrow keys to select Dark mode (or whichever you prefer) and hit Enter. It’s just asking how you want the terminal to look – purely cosmetic.

After this, Claude Code asks you to authenticate with your Anthropic account. Same account you use for Cowork. Follow the prompts – takes about 30 seconds.

5. Recommended settings

You’ll get a popup asking about your settings – click Yes, use recommended settings and hit Enter. It’s just optimizing how the terminal displays text for Claude Code – you want this on.

6. You’re in

Here is what it will look like:

7. Your first task

Before configuring anything else, try something. If you have a CSV, a research doc, or a set of call transcripts in your folder, type:

Read [filename] and tell me what's interesting. Any patterns, any surprises, anything that stands out.

Claude reads the actual file on your computer and comes back with analysis. No uploading. No copy-pasting into a chat window. It just went and looked – the same way Cowork does, but from the terminal.

The Cowork-to-Code bridge

If you’ve been following this series, you already have most of what you need.

Your Skills transfer

Every Skill you built or installed – the PMM Skills Pack, your custom positioning Skill, your battle card Skill – works in Claude Code exactly the same way. Same SKILL.md files, same automatic detection, same methodology.

Your context files transfer

If you set up context files in your Cowork folder (about-me.md, voice-and-style.md, working-rules.md from Part 2), Claude Code reads those too. Point Code at the same folder and it already knows who you are.

CLAUDE.md replaces Global Instructions

In Cowork, you set persistent preferences in Settings –> Global Instructions. In Code, the equivalent is a CLAUDE.md file.

There are two layers: a global one at ~/.claude/CLAUDE.md that applies everywhere, and a project-specific one in whatever folder you’re working from. Same concept, same purpose – Claude reads it automatically at the start of every session.

If you already have Cowork’s Global Instructions dialed in, copy them into your global CLAUDE.md file. Your preferences carry over.

The briefing model is the same

Describe the end state. Provide the inputs. Set the constraints. Delegate. The only difference is that Code can do more with the brief – it can build, install, connect, automate, and run things that Cowork can’t.

The jump is smaller than it sounds. You already know how to brief Claude effectively. That’s the hard part. The terminal is just a different window.

How to talk to Claude Code

A few controls worth knowing if you’re new to the terminal.

Permission modes. Claude Code asks permission before taking actions. Hit Shift+Tab to cycle through three modes:

• Default: Claude asks before every action. Start here. Stay here until you’re comfortable.
• Auto-accept edits: Claude edits files freely but checks before running commands. Good once you trust it with your working folder.
• Plan mode: Claude goes read-only. It analyzes everything, builds a plan, but changes nothing until you approve. Type /plan to jump straight into this. Great for complex tasks where you want to see the roadmap first.

Context management. Every conversation has a limit – think of it as a glass filling up. Every message, every file Claude reads, every response uses some of that capacity.

•  /context shows how full the glass is.
•  /compact compresses the conversation to free up space.
•  /clear starts fresh.

The golden rule: one task per conversation. Don’t ask Claude to build a dashboard and then draft an email and then analyze a spreadsheet in the same session. Each unrelated task degrades the quality of everything else. Start fresh for each major task.

@filename pulls a specific file into the conversation. Instead of describing where your positioning doc lives, type @positioning-v3.md and Claude reads it directly.

ESC stops Claude mid-response. Hit it the moment you see it heading somewhere wrong. ESC twice opens the rewind interface – a time machine that lets you roll back to any point in the conversation.

Where Code changes PMM work

Code gives PMMs three capabilities that Cowork can’t: interactive outputs, automated workflows, and analysis at scale. Each one maps to work you’re already doing – with a fundamentally different type of output.

I’ll walk through each with the actual steps so you can try them.

1. Interactive outputs

Cowork produces documents. Code produces things people interact with – pages, dashboards, tools that open in a browser and let you click, filter, and explore.

Turning a product brief into a launch page prototype

The situation: your product manager ships a brief for a new feature. You need buy-in on the positioning before the launch page gets designed and built. Normally, the options are a Google Doc with bullet points or waiting two weeks for design bandwidth. Code gives you a third option.

How to do it:

1. Save your product brief and your messaging framework into your working folder (or the /inputs subfolder if you set one up in Part 2).
2. Open Claude Code in that folder.
3. Type something like this:

Read product-brief-analytics.md and messaging-framework-v2.md. Build me a working prototype of the launch page. I need a hero section with headline and subhead, three sections explaining the key value props, a comparison table showing us vs the two main competitors, and a signup button at the bottom. Use these brand colors: [your hex codes]. Make sure it looks good on mobile.

4. Claude reads both files, then starts writing code. You’ll see lines scrolling in the terminal – that’s Claude building the page. Let it run. Takes a couple of minutes.
5. When it finishes, it tells you it created a file – something like analytics-launch-page.html. Go to your folder, find it, double-click it. It opens in your browser.

You’re now looking at a fully functional landing page prototype. Interactive. Responsive. Built from your actual messaging and your actual product brief.

Walk into the positioning review with this instead of a slide deck. Stakeholders react differently when they can see the messaging in context on a real page – not as bullet points in a doc.

The prototype isn’t production code. Your engineering team won’t ship it as-is. But it’s a communication tool, and that’s what the job is.

Want to test three different positioning angles? Tell Claude: “Now build me two alternative versions – one that leads with the speed benefit and one that leads with the integration story.” Compare all three side by side in your browser. Ten minutes, three prototypes, a clear recommendation.

Win/loss dashboards

The situation: you’ve got a spreadsheet of closed deals – won and lost – with fields like competitor mentioned, deal size, buyer persona, stage lost, and loss reason. In Cowork, you’d get a document summarizing the patterns. Useful, but static. You send it once, nobody opens it again.

How to do it:

1. Export your deal data as a CSV from your CRM (HubSpot, Salesforce, whatever you use). Save it to your working folder.
2. Open Claude Code in that folder.
3. Type:

Read deal-data-q1.csv. Build me an interactive win/loss dashboard. I need: win rate by competitor, win rate by buyer persona, the most common loss reasons ranked by frequency, average deal size for wins vs losses, and filters so I can drill down by time period and by competitor. Save it as an HTML file.

4. Claude reads the data, builds the dashboard, saves it. Double-click the file to open it in your browser.
5. Click through the filters. See which competitor you’re losing to most. See which persona converts best. Share the file with your sales leader – they open it in Chrome and explore it themselves.

When Q2 data comes in, drop the updated CSV in the same folder and tell Claude to refresh the dashboard. Same filters, updated numbers.

I literally built this interactive dashboard (with example data) in under 1 minute. Honestly mind-blowing.

Messaging testing tool

The situation: you’ve got three positioning options for a new feature and need structured feedback from five stakeholders before the launch brief is due.

How to do it:

1. Save your three messaging options as a document in your working folder (or just have them ready to paste).
2. Open Claude Code and type:

I have three positioning options for our new analytics feature. Build me a feedback tool - an HTML page that shows all three versions side by side, with a vote button and a comment box under each option. Add a summary section at the top explaining what I'm asking people to evaluate. Save it as messaging-test.html.

3. Send the file to your five stakeholders. They open it in their browser, read the options, vote, and leave comments.
4. Structured feedback, collected asynchronously, without scheduling a meeting.

2. Automated workflows

Cowork can schedule tasks (we covered this in Part 2). Code goes further – it builds the underlying system that makes the automation work: the logic that compares this week’s data to last week’s, the rules for what counts as a meaningful change, the file organization that keeps everything clean over time.

Competitive monitoring

The situation: you need to track what your competitors are doing with their pricing, features, and messaging – but the last competitive update you sent to sales was two months ago because doing it manually takes a full day.

How to do it:

1. Open Claude Code in your working folder.
2. Type:

Build me a competitive monitoring system. I need it to check these three websites every Monday at 9am: [competitor1.com], [competitor2.com], [competitor3.com]. For each one, capture their pricing page, their features page, and their homepage messaging. Compare what's there now against what was there last week. If anything changed, flag it. Save a weekly digest as a markdown file in a /reports folder with the date in the filename. If anything significant changed, also create a short summary I can share with the sales team.

3. Claude builds the whole system – the script that checks the sites, the scheduling, the folder structure, the comparison logic. It stores baseline snapshots of each page so it has something to compare against next week.
4. Every Monday, a competitive digest appears in your /reports folder. Open it, scan for changes, forward the summary to sales if anything meaningful shifted.

The monitoring runs automatically as long as your machine is on and Claude Desktop is open when the schedule triggers. If your machine is asleep on Monday at 9am, that run gets skipped – it picks up the following Monday.

Launch readiness tracker

The situation: your launch has 40 checklist items across product, sales, marketing, and CS. You’re currently tracking this in a spreadsheet that three people forget to update.

How to do it:

1. Write your launch checklist in a simple text file or spreadsheet – task, owner, workstream, status.
2. Open Claude Code and type:

Read launch-checklist.csv. Build me an interactive launch tracker - group the items by workstream, show the owner for each task, add a status toggle (not started / in progress / done), and show a progress bar for each workstream at the top. Save it as launch-tracker.html.

3. Claude builds the tracker. Open it in your browser. Share the file with your launch team.
4. Each day, open Claude Code and tell it what’s changed: “Mark the sales enablement deck as done. Move the pricing page update to in progress. Add a new item: CS training session, owned by Sarah, under customer success.” Claude updates the file.

Sales objection library

The situation: your sales team keeps asking for the same competitive talk tracks in Slack. You’ve answered the same question about your pricing vs. Competitor X at least four times this quarter.

How to do it:

1. Export your call recordings data from Gong or Chorus (most tools let you export transcripts or summaries as CSV). Save it to your working folder.
2. Open Claude Code and type:

Read gong-export-q1.csv. Build me a searchable sales objection library. Organize the objections by category (pricing, features, security, implementation, etc). For each objection, show the most effective responses from calls we won, and tag the win rate for each response pattern. Make it searchable so reps can find what they need fast. Save as objection-library.html.

3. Share the file with your sales team. They bookmark it. When a rep hears “your competitor is cheaper” mid-call, they search the library and find the talk track that’s worked before – with the data to back it up.
4. When new call data comes in next quarter, drop the updated export in the folder and tell Claude to refresh the library.

3. Analysis at scale

Cowork handles individual files well. Code handles dozens of files simultaneously and can cross-reference across them – which is where PMM research gets powerful.

Multi-source market analysis

The situation: you’re preparing a board-level market review. The inputs are scattered across five different formats: 12 customer interview transcripts, a competitor feature comparison spreadsheet, three analyst reports (PDFs), internal product usage data (CSV), and your current positioning doc.

In Cowork, this would take multiple sessions and a lot of manual stitching. Code handles it in one.

How to do it:

1. Create a subfolder in your working directory – something like /board-prep. Drop all the files in there: the transcripts, the spreadsheet, the PDFs, the usage data, the positioning doc.
2. Open Claude Code in your working folder and type:

Read everything in /board-prep/. I need a market analysis covering four areas: customer sentiment trends from the interview transcripts, competitive positioning gaps from the comparison spreadsheet and analyst reports, product-market fit signals from the usage data, and where our current positioning aligns or conflicts with all of the above. Build it as an interactive HTML report with clickable sections for each area. Also save a plain text summary I can paste into the board deck.

3. Claude reads every file, cross-references across sources, and produces both outputs: the interactive report for your deep dive and the summary for the deck.

One prompt. Five different file types. Two output formats. The kind of synthesis that used to take a week of reading, highlighting, and manually connecting dots.

Voice-of-customer synthesis

The situation: you’ve run 20 customer interviews across three segments over two months. The transcripts are sitting in a folder. You need to turn them into a structured research report with themes, evidence, and implications for your messaging.

How to do it:

1. Save all 20 transcripts into a subfolder (e.g., /customer-interviews).
2. Open Claude Code and type:

Read all the transcripts in /customer-interviews/. Map the recurring themes by customer segment. For each theme, pull the strongest verbatim quotes as evidence. Identify where the language customers actually use aligns or conflicts with our current messaging in messaging-framework-v2.md. Produce a structured voice-of-customer report with themes ranked by frequency, organized by segment.

3. Claude reads all 20 files, identifies patterns across them, and produces a report grounded in the actual data. The kind of qualitative analysis that used to take a full week of coding and tagging – done in one session.

Pipeline pattern analysis

The situation: you want to understand why you win or lose against specific competitors, which buyer personas close fastest, and where deals stall in the pipeline. The data exists, but it’s spread across three tools.

How to do it:

1. Export six months of deal data from HubSpot (CSV). Export relevant competitive mentions from Slack (you can use Claude Code’s Slack connector for this). Export feature adoption data from Amplitude (CSV).
2. Drop all three exports into your working folder.
3. Open Claude Code and type:

Read deal-data.csv, slack-competitive-mentions.csv, and amplitude-usage.csv. I want to understand the patterns: which product features correlate with wins against specific competitors? Which buyer personas close fastest and at the highest deal sizes? Where in the pipeline do deals stall and what are the common reasons? Build me an interactive dashboard that lets me explore these patterns with filters for competitor, persona, and time period.

4. Claude reads all three files, looks for correlations across them, and builds a dashboard. The cross-referencing across data sources is the part that’s hard to do manually and expensive to request from a BI team. Here, it’s one prompt.

Connecting to external tools

In Part 4, we covered Connectors – one-click integrations you set up in Cowork’s settings. Slack, HubSpot, Google Drive, Amplitude, and dozens more.

Those same connectors work in Code too. Connect once, available everywhere.

Code also lets you connect to tools that aren’t in the standard directory.

If your company has an internal database, a proprietary CRM, or an analytics tool that doesn’t show up in Cowork’s connector list, Code can likely reach it through something called an MCP server (think of it as a custom bridge between Claude and the tool).

The setup is a one-time install – Claude walks you through it when you ask, and after that it loads automatically every session.

Worth flagging to your ops or engineering team as a future project if there’s a data source Claude should be able to access but can’t.

When to use Code vs. Cowork

You don’t need to choose one. They complement each other. Here’s the decision framework:

Use Cowork when:

• You need a document: a battle card, a positioning doc, a launch brief, a research summary
• The task is self-contained: one brief in, one deliverable out
• You want the desktop app experience: visual file management, drag and drop, browser control built in
• You’re producing something for stakeholders to read

Use Code when:

• You need something interactive: a dashboard, a prototype, a tool
• You need automation: monitoring, scheduled reports, recurring workflows
• You need to work across many files simultaneously: research synthesis across 20+ documents
• You need to connect to tools outside the standard connector directory
• You’re building something your team will use repeatedly

The simplest test: if the output is a file someone reads, Cowork. If the output is a tool someone uses, Code.

And remember – your Skills and context work in both. The methodology you’ve built doesn’t change. The surface does.

Honest limitations

Worth knowing upfront rather than discovering mid-project.

The terminal is unfamiliar

There’s no way around this. If you’ve never opened a terminal before, the first session feels alien. The learning curve is real, but it’s shorter than you expect.

The PMMs in our community who’ve tried it tend to say the same thing: they were comfortable within a few sessions. The interaction model (typing sentences, reading responses) is the same as Chat and Cowork. The window it happens in is different.

No visual file management

Cowork shows you your files in a sidebar. Code shows you text in a terminal. You can ask Claude to list files, and you’ll get comfortable with it quickly, but it’s a different way of working. Some people prefer it. Some don’t.

No memory across sessions

Same constraint as Cowork. Claude doesn’t remember previous conversations. The handover file practice from Part 2 applies here too – ask Claude to write a HANDOVER.md at the end of complex sessions. Start the next session with: “Read HANDOVER.md and continue from where we left off.”

Your machine needs to be running

Same as Cowork. If your computer is asleep or the terminal is closed, nothing executes. Scheduled automations only run when your machine is awake.

Built outputs need sharing for team access

The dashboards and tools Claude builds are files on your computer. They open in any browser – but they live on your machine, not on the internet.

For your team to use them, you need to share the files: put them in a shared Dropbox or Google Drive folder, email them directly, or host them internally. They’re not websites with a public link (though Claude can help you set that up if you go that route).

It can break things if you’re not careful

Claude Code has real access to your file system. It can create, edit, move, and delete files. Start in default permission mode (asks before every action).

Keep your working folder dedicated – don’t launch Code from your home directory. And back up important files before any major task. The same rules from the Cowork security section in Part 2 apply here, arguably more so.

The bottom line

In Part 1, I said knowing Claude Code exists changes how you think about what’s possible. After pressure-testing it, I’d go further: Claude Code is where PMMs should go once they’ve outgrown documents.

Chat gave you a thinking partner. Cowork gave you a production engine. Skills gave you encoded methodology. The marketplace gave you connected tools.

Code gives you the ability to build. Dashboards, prototypes, monitoring systems, internal tools – things that used to require a developer, a ticket, and a three-week wait.

If you’ve set up Cowork and built your Skills, you’re already 80% of the way there. The last 20% is opening a terminal and trying one of the workflows above. Start with whichever one solves a problem you actually have this week.

The full series: Mastering Claude as a product marketer

Part 1: Mastering Claude for product marketing – What everything actually is. The map.

Part 2: Claude Cowork for product marketing – How to set it up, the PMM workflows that benefit most, and how to go from “chat responses” to “finished deliverables in your folder.”

Part 3: The complete product marketing Claude skills pack – Everything you need to master Claude skills for product marketing

Part 4: The Claude marketplace for product marketing – Skills, plugins, and connectors. How to find them, install them, and build your own PMM toolkit.

Part 6: Bringing it all together – Claude for product marketing – How Chat, Cowork, Code, Skills, and Connectors work as one system. The full PMM Claude setup, from first install to daily workflow.