The marketer's case for Claude Code

Chat vs. Cowork vs. Code: there is no spoon

Baja California Sur, México. Photo by author David E. Sweenor

Skip the Claude mode debate

Most marketing teams I’ve worked with this past year are running similar setups. ChatGPT or Gemini lives in a permanently open browser tab, and the team’s saved prompts live in a Slack channel or Google doc that they copy and paste from. That’s a great way to get started, but as usage matures, the team needs more coordination and better tooling. Teams have seen all of the hoopla around Claude Code, and many of my clients are moving in that direction. Inevitably, they always end up asking the same question. How do I get started with Claude and which mode (Chat, Cowork, or Code) should I be using?

Take a gander around Substack and you’ll see an endless debate about which mode to use when. Chat for quick questions, Cowork to help connect your files and tools like Gmail, and Code for the technical bits. My recommendation is the same one I follow myself. Use only Claude Code. That’s it.

Remember the movie The Matrix? This is the same red pill / blue pill choice that Neo had. The browser tab is the blue pill. It’s easy and familiar, every conversation is a fresh start, and nothing compounds. Claude Code is the red pill. The editor is unfamiliar for a day. After that, you can’t unsee what’s on the other side and there’s no turning back.

The install takes 15 minutes, the vocabulary another 10. This post covers both, plus the case for skipping the debate.

A tool is not a stack

Installing Claude doesn’t change the AI itself. The model in your browser tab and the model in Claude Code are the same Anthropic model. But, the wrapping and context around the model differ. The additional context you build inside Claude Code turns the AI from a tool into a full-fledged marketing operating system. You can use it on your own, or, as your team scales, you can share context across your entire marketing or GTM function.

Browser-tab AI gives you the model and a chat window. That’s it. Memory exists, but it’s behind a separate settings screen, and you update it manually. The memory itself may leak across contexts you didn’t ask it to span. When you advise multiple clients or work on anything sensitive like product roadmaps, acquisitions, and launch plans, that can be problematic to say the least. The conversation you had about your upcoming product launch is one ambient prompt away from showing up in your conversation or content that’s being pushed to public channels before you’re ready. Consultants don’t want info from Client A leaking into Client B. You have better control over this in Claude Code.

Chat-only setups don’t share very well. Your colleague can’t inherit the prompts and context you’ve built up. Every team conversation is like a Saturday Night Live (SNL) cold open. Claude Code gives you the model plus four pieces that surround it that are transformative..

Memory you can scope to a project, so what Claude knows about Client A stays with Client A.
Skills you write once and reuse across the team with a slash command.
CLAUDE.md is a markdown file the team can read and edit, so the brand voice and deliverable formats are the same whether you’re drafting or your colleague is.
MCPs wire Claude into the tools you already use, like Gmail, Sheets, and Calendar. You don’t switch tabs or copy data into a chat. Claude reaches into the tools where the data lives.

Skills, CLAUDE.md, and MCPs all live in Code mode. Memory follows you across modes, and Code lets you scope it per project. That’s not coincidence. It’s why the rest of this post leans toward Cursor.

The 15-minute install

Four installs, about 15 minutes total. You can do this between meetings. I’ve run this install on my Mac and helped clients on Windows machines, and it doesn’t take long. To get started, follow these steps:

Start with a free GitHub account (2 minutes). This is where your team’s repo lives. Have the repo owner add you and accept the invite when it lands in your inbox.
Next, download Cursor and sign in with your work account (5 minutes). Cursor is the workspace where Claude Code runs. Defaults are fine and there is no need to buy a subscription.
Then install the Claude desktop app (3 minutes). You’ll mostly live in Code, but the desktop app is where you manage your account, and it pairs with Claude Code on the same login.
Finally, install Claude Code itself (5 minutes). Open Cursor’s terminal (View → Terminal) and paste the install command for your OS. On macOS or Linux, that’s curl -fsSL https://claude.ai/install.sh | bash. On Windows PowerShell, it’s irm https://claude.ai/install.ps1 | iex. Type claude and sign in when prompted. Claude Code requires a paid Anthropic plan (Pro, Max, Team, or Enterprise).

Here’s the checklist:

Open Cursor and sign in
Open the Claude desktop app and sign in
Run claude in Cursor’s terminal without errors
See your team’s repo on your GitHub account

When you get stuck, ask Claude. It can walk you through any install error, terminal command, or path issue you hit.

The three modes (and the one you need)

“Chat answers. Cowork connects to your tools. Code does all of that and more.”

— David Sweenor, Founder/CEO, TinyTechGuides

You’ll see three modes in the docs. Here’s what each one does, then what to do with

Chat: Back-and-forth conversation in a window. Best for quick questions, drafts, and brainstorming.
Cowork: Claude plans and executes a task on your files, apps, and browser. Best for multi-step work that doesn’t live in a repo.
Code: Terminal-based, runs inside a repo. Best for shared team docs, branded decks, and anything you want versioned.

Each mode has its own use case in the docs and its own corner of the marketing AI internet defending it. Last week I saw a debate about whether spreadsheet research belongs in Cowork or Code. The week before, someone argued that brainstorming should never happen anywhere except Chat.

Each row in that table is technically true, but none of them gets you to a marketing OS. Code does.

Code is where Skills, CLAUDE.md, and MCPs live. MCP stands for Model Context Protocol, the way Claude connects to other systems like Gmail, Asana, Canva, and Calendar. Code is where work compounds across sessions, where memory gets project-scoped, and where the team can see and edit what you made (via GitHub).

The most underrated benefit shows up when you need something none of those four pieces provide. An action-item tracker that pulls from your meeting notes. A status-report rollup across all your clients. Or a one-off Python script to clean up your Downloads folder. Code writes that on the fly. Browser-tab AI can’t write code that touches your files. Code can.

Teams matter too. When you ship work in Code to GitHub, your colleague clones the repo and inherits everything you’ve built. The Skills you wrote. The CLAUDE.md that defines the brand voice. The project-scoped memory rules. Browser-tab AI doesn’t ship. Your work stays on your account, where no one else can use it.

The Cursor interface is unfamiliar for the first day. The first 30 seconds are the worst. There’s a sidebar, a file tree, and a terminal pane. Push through it. After that, you’ve crossed the line, and you don’t go back, ever.

The four components that make Claude Code so powerful

Four components configure Claude. Three define how Claude works. The fourth wires Claude into the tools where your work already lives.

Memory is about you. Your role, your voice preferences, and how you like content formatted. Personal and portable across every project you work in. You set it once, and Claude carries it from session to session. After three months of corrections, the memory you’ve built is what makes Claude sound like you instead of a smart intern.

Skills are about the task. A Skill is a reusable recipe (instructions plus examples) that you call with a slash command. /write-blog. /generate-deck. /prep-podcast. Each one bundles the steps Claude should take for that task and the format of the output. For anyone already writing prompt workflows, Skills are how those workflows become team-shareable assets in your repo. Write a Skill once, and it works the same way every time you call it.

CLAUDE.md is about the project. It’s a markdown file in the project’s repo that Claude reads every time it works in that scope. It’s where you write down what’s true for this project. Brand voice. Audience. Deliverable formats. Drop one in your TTG repo, drop another in your client repo, and Claude switches voices without you reminding it.

MCPs wire Claude into the tools where your work already lives. Where Memory, Skills, and CLAUDE.md tell Claude how to work, MCPs tell Claude where to reach. Each MCP connects Claude to a specific tool, like Gmail, Sheets, Calendar, or Asana. You don’t have to copy data into a chat. Install one and Claude reaches into that tool. Install several and Claude works across your whole toolchain.

“Memory is about you. Skills are about the task. CLAUDE.md is about the project, and MCPs connect your apps.”

— David Sweenor, Founder/CEO, TinyTechGuides

A growing set of community patterns extends the same idea. A popular one is napkin.md, a per-repo runbook that Claude reads at session start and curates as you work. It catches mistakes once and stops repeating them. You commit it or you don’t, your call.

Your first move

You finished the install. Now what?

Open Cursor’s terminal, run claude, and have a five-minute conversation. Tell Claude who you are. Tell it what you do. Tell it the three things you reach for most often in a week. Save what you tell it as memory.

That’s the first move. Anything more is week-two work. Your first Skill conversion, your first CLAUDE.md, and your first MCP can wait. None of that matters in week one. Build the habit of opening Cursor, typing claude, and treating it like a colleague who started Monday.

Take the red pill

Fifteen minutes is the easy part. The IDE filters out most marketers. The marketing OS rewards the ones who push through.

Most marketers won’t cross the gate, which is why their AI output never compounds. You’re reading this, which means you’re closer than you think. Cross the gate. Take the red pill.

This post is the entry point for the Claude for Marketing Operators series on TinyTechGuides. The next eight posts walk through each layer of your marketing OS. The four-layer mental model. CLAUDE.md teardowns from real client projects. The MCPs that pull their weight, and the ones that don’t. What six months of memory looks like across multiple clients. How a solo consultant runs four engagements on one brain. Why your marketing OS becomes the moat your competitors can’t copy.

If you want each one as it drops, subscribe to the Marketing section on Substack.

Frequently asked questions

What is Claude Code?

Claude Code is the terminal-based mode of Anthropic’s Claude AI assistant, built for work that lives in a code repository. For marketers, it runs inside an editor like Cursor and lets you maintain a project rulebook (CLAUDE.md), reusable Skills, scoped memory, and connectors (MCPs) to tools like Gmail and Sheets. It’s also the only mode that can write Python on the fly to handle tasks none of those four pieces cover, like building an action-item tracker or a status-report rollup across clients.

How does Claude Code differ from Chat and Cowork?

Chat is back-and-forth conversation. Cowork lets Claude plan and execute multi-step tasks across your files, apps, and browser. Code runs in your terminal inside a code repository, so the work compounds across sessions through Skills, CLAUDE.md, scoped memory, and MCPs. Chat and Cowork are useful for narrow tasks. Code is where reusable assets live and where teams can share what one person built. For marketing operators using Claude in production, the recommendation is to skip the mode debate and run everything in Code.

Do I need to be a developer to use Claude Code?

No. Claude Code runs in a terminal, but you don’t need to write code to use it. You install it once, then talk to Claude in plain English. The terminal interface is unfamiliar for the first day, but most marketing tasks (drafting blog posts, building decks, summarizing client meetings, packaging prompts as reusable Skills) work the same way they do in a chat window. The benefit is that everything you do compounds across sessions and can be shared with your team through a repo.

Does Claude Code require a paid Anthropic plan?

Yes. Claude Code requires a Pro, Max, Team, or Enterprise plan from Anthropic. The free Claude.ai plan does not include Claude Code access. The desktop app, browser-based Claude.ai, Cursor, and GitHub are all free — only the Claude Code CLI itself needs a paid plan. You can install all four pieces (GitHub, Cursor, Claude desktop app, Claude Code) before deciding on a plan, since the install completes before authentication.

What’s the difference between memory and a CLAUDE.md file?

Memory is personal and portable across every project you work in: your role, voice preferences, and how you like content formatted. CLAUDE.md is project-specific. It’s a markdown file in a project’s repo that Claude reads every time it works in that scope, capturing brand voice, audience, deliverable formats, and anything else that’s true for that project. Memory follows you across modes; CLAUDE.md stays with the project. A solo consultant runs one memory and many CLAUDE.md files, one per client.

Where should I start if I want to use Claude for marketing work?

Install the four pieces (GitHub, Cursor, Claude desktop app, Claude Code) in about 15 minutes. Then open Cursor’s terminal, run Claude, and have a five-minute conversation. Tell Claude who you are, what you do, and the three things you reach for most often in a week. Save what you tell it as memory. That’s the entire first move. Skill conversion, CLAUDE.md drafting, and MCP setup are all week-two work. Week one is opening the terminal and starting to use it.

About David Sweenor

David Sweenor is the founder and host of the Data Faces podcast, where he talks with the people who are making data, analytics, AI, and marketing work in the real world. He is also the founder of TinyTechGuides and a recognized top 25 analytics thought leader and international speaker who specializes in practical business applications of artificial intelligence and advanced analytics.

With over 25 years of hands-on experience implementing AI and analytics solutions, David has supported organizations including Alation, Alteryx, TIBCO, SAS, IBM, Dell, and Quest. His work spans marketing leadership, analytics implementation, and specialized expertise in AI, machine learning, data science, IoT, and business intelligence. David holds several patents and consistently delivers insights that bridge technical capabilities with business value.

Books

Follow David on Twitter @DavidSweenor and connect with him on LinkedIn.