The Marketer's Guide to Building with Claude Code
The exact process I use to go from idea to a live app. No computer science degree required. What to build, how to set up, real cost numbers, and case studies from things I've actually shipped.
By far the biggest unlock I've had this year was getting comfortable with Claude Code. It took me from idea to a live product in record time — and I'm not a developer. I don't write code. I describe what I want, and I direct the thing that builds it.
It's also the question I get asked more than almost any other: how do I actually start?
This is the answer — the exact process I use to build agents, apps, and tools. No theory. What to build, how to set up, the mental model shift you have to make, common failure modes, real costs, and five case studies from things I've shipped that this site links to.
What Claude Code actually is
Most people conflate Claude Code with Claude the chatbot. They're related but different tools for different jobs.
Claude.ai (the chat) is where you talk to Claude to think, brainstorm, draft, and explore. Best for one-shot output: a summary, a rewrite, a plan, a Custom GPT config, a document.
Claude Cowork is Anthropic's desktop app — Claude running as a native application on your computer, with access to files across your system rather than a single project folder. It sits between claude.ai (browser, one-shot) and Claude Code (terminal, deep in a project). Best for ongoing side-of-the-screen work that doesn't belong in a specific codebase — organizing files, analyzing across documents, or having Claude nearby while you work in other apps.
Claude Code is Claude running inside your terminal, with access to your files and your project. It reads the code, writes new code, runs commands, tests changes, and shows you what it did before applying. It's the difference between describing a car in a chat and directing a mechanic who's standing in your garage. This guide is about Claude Code.
What you can actually build without being a developer
The projects worth building fall into five buckets, roughly ordered by complexity. Pick one that fits your current time budget and grow into the next.
A small tool that solves an annoyance in your life. My QR card generator is this. Built because typing my name into people’s phones after events got old.
A brief intake form. A prompt library your team can search. A dashboard viewer. Anything that used to be a "someday" idea because building it required a developer.
A scorecard, a calculator, a use-case finder. My AI Enablement Toolkit is four of these. They generate inbound, prove your POV, and cost almost nothing to run.
A personal site with articles, tools, and identity built in. The site you’re reading right now is this. It gets refined every week and I never touched a docs page.
A real thing with real users. MediaPlan.ca is this. Not a hobby project — a working operating system for media planners that sustains monthly iteration.
The mental model shift
The biggest gap between non-coders who ship and non-coders who stall isn't technical. It's how they think about the work. Four shifts that make everything easier:
Don’t say "add a button that does X." Say "when someone finishes the scorecard, they should be able to share their score with their team." Claude Code figures out the button. You figure out what "done" looks like.
You don’t need the whole thing designed. Start with the smallest working version. Iterate five times in an afternoon. That’s the loop that produces good software — for developers too.
Claude Code shows you what it changed before applying it. Read those changes. If something looks wrong, ask for a revert. This is the single habit that separates people who stay in control from people who lose weeks to weird bugs.
"Make it look nicer" produces generic output. "Match the visual style of the section above it, use the same border radius and hover state" produces exactly what you meant. Specificity in equals specificity out.
Set up your workshop
The first four steps are just setting up the room you'll work in. Give yourself an hour the first time.
Pro, Max, Teams, or Enterprise — any tier unlocks Claude Code. If you're going to build regularly, Max at $200/month is the tier I'd recommend. It's the difference between hitting usage limits mid-project and being able to build for hours without thinking about it.
It's the editor I run everything through. You don't need to understand it in any deep way. The reason I like it: I can see my whole file structure on the left while I work, which makes the project feel real instead of like a black box.
Open your terminal (Terminal on macOS, PowerShell on Windows) and paste this command:
This installs Claude Code globally on your machine. If you don't have Node.js installed, install that first from nodejs.org ↗ (grab the LTS version).
To verify the install worked, run:
You should see a version number print. If you do, you're installed. The first time you run claudeinside a project, it'll open a browser window and ask you to sign in with your Anthropic account. That's the one-time authentication step.
Empty folder on your desktop. Name it after the project you want to build. Open it in VS Code. Then type claude in the terminal. That launches Claude Code right inside your project.
Now you're ready to build.
The building loop
There are two ways to start a project, and I use both depending on how clear the idea is in my head.
Method 1: Describe it directly in the terminal
Type your project description into Claude Code and let it enter plan mode. It maps out what it's going to build before it builds it. You approve the plan, or push back on it, before any file gets created. This is the most disciplined way to start — it forces the thinking upfront.
Method 2: Spec it in claude.ai first
This is the shortcut I lean on. Before I open Claude Code, I chat the idea through on claude.ai — describe the audience, what it does, the flow, the constraints, the visual style. I get a fully fleshed-out spec. Then I ask claude.ai to package all the files I need to get started. I drop those into my project folder and have Claude Code review them and implement the plan.
It's a small trick that saves real time. Claude.ai is better at ideation; Claude Code is better at implementation. Use each for what it's good at.
Either way, the pattern from there is the same:
You describe what you want next.
Claude Code proposes changes.
You read the diffs before approving.
It applies them.
You test the change (open the app in your browser, click through it).
You iterate: what worked, what didn’t, what to tweak.
You bring the thinking. It brings the building. The loop closes in minutes instead of days.
Put it somewhere real
A project living only on your laptop is fragile. Two steps fix that.
Private repo for personal projects. Ask Claude Code to upload your project to it. This does two quiet but important things: it backs your work up to the cloud, and it gives you version control — nothing you build can be lost or broken beyond recovery.
Vercel publishes your project to a live URL. If that sounds intimidating, it isn't. Claude Code walks you through connecting the two, step by step. A few minutes later, the thing you built has a real web address.
From now on, every time you make a change in Claude Code, you just ask it to push to GitHub, and Vercel automatically picks it up and updates your live URL. You edit, you ask it to ship, and the public version is always current. The loop closes itself.
Common failure modes
Everyone who builds this way runs into the same six problems in the first month. Naming them makes them cheaper.
Trying to design every feature upfront before writing any code. You freeze. The fix: write a two-line spec, build the smallest possible working version, iterate from there.
Vague prompts like "build me a marketing tool." You get generic output because that’s the only signal you gave. The fix: describe audience, purpose, constraints — the same discipline you’d use for a creative brief.
Claude Code shows you every file it wants to modify. Skimming past the diff and clicking approve is how projects break in subtle ways. Read the changes, especially in the early days.
If you’re not pushing to GitHub regularly, you’re one bad edit away from losing hours of work. Push after every meaningful change — ideally at the end of every session.
"While you’re at it, also add..." — the phrase that turns a two-hour project into a two-month project. Ship the smallest useful version, then decide what version 2 needs.
Claude Code is for building projects. For a one-shot rewrite of a paragraph, a Custom GPT config, or exploring an idea — claude.ai chat is better. Don’t reach for Claude Code for everything.
What it actually costs
The honest breakdown.
$200/month for Claude Max (20x tier)is what I run. There's a $100 Max tier (5x usage) and Claude Pro at $20/month works for lighter usage, but Max is what makes multi-hour building sessions comfortable. Claude Code is included in these — no separate pricing.
$0 for most personal projects.Vercel's free tier covers everything until you have real traffic. GitHub is free for private repos. You only start paying when a project has meaningful users or needs specific services (a database, a paid API).
Domain (~$15/year), a transactional email service like Resend (~$20/month if you send meaningfully), model API costs if your app calls Claude directly (varies wildly). Most of my projects run under $50/month all-in once live.
For context: I've shipped five substantial builds this year for less than the cost of what a freelance developer would charge for one of them.
Five things I've built with this
Everything I've shipped this year came out of the same setup and the same loop. Same subscription, same VS Code, same push-to-Vercel pipeline. Different projects, different scopes.
The most ambitious thing I’ve built with Claude Code. An AI-powered operating system for media planners — brief intake, plan generation, flowchart, competitive layer. Real users, real workflow. What I learned: shipping small versions weekly beats designing the perfect product for months. Every version taught me something no upfront plan would have.
The scorecard, the ROI calculator, the use-case finder, the prompt library — all on this site. What I learned: interactive tools need 3x the polish of static content. The scorecard alone went through eight visual iterations before it felt right. Claude Code made those iterations cost 30 minutes each instead of a full weekend.
Built after SocialWest 2026 because typing my name into people’s phones after every conversation got old. Context-aware from the start — a QR that adapts to where it’s scanned. What I learned: the fastest way to know if something’s useful is to build it in a weekend and use it the next week.
This site. Articles, workflow guides, tools, speaking, about, footer, the chat widget, the whole thing. Every section got shaped by feedback and use. What I learned: building in public compounds — every page I ship makes shipping the next page cheaper because I have more patterns to reuse.
Custom tools, briefs, and small agents that make specific parts of paid media work faster. What I learned: the biggest wins aren’t the flashy projects. They’re the small internal tools that used to sit on the "someday, when we have a developer" pile. Now those ship in a day, and they save more hours than the big ones.
Where to start, by where you are now
The right first project depends on how much time you have and where your comfort level is. Pick the smallest one that scares you a little.
Start with a personal utility. A weekend, maybe less. Pick a small annoyance in your life and build the thing that fixes it. My QR card is this. It doesn’t matter if anyone else ever uses it — the point is finishing something and shipping it live.
Build an internal tool for your team. A day or two. A brief intake form, a prompt library, a simple dashboard. The bar is: something one other person on your team would use once a week.
Build a client-facing free tool. A week of evenings. A scorecard, a calculator, a use-case finder. Something that puts your POV in front of strangers and generates inbound.
Ship something ambitious. A month of focused effort. A publication, a real product, an agent that runs on a schedule. The kind of project that used to require hiring someone.
Where this fits
This piece is the practical how-to under the broader “build things teams will use” theme. Three adjacent pieces worth reading alongside it:
My broader stack — the tools I use alongside Claude Code, and how they fit together.
The broader framework — where building tools fits inside a marketing team's AI workflow.
Why every knowledge worker learning to ship tools this way matters at the org level.
The tools this guide references: MediaPlan.ca ↗, the AI Enablement Toolkit, the QR Card Generator, and this site itself. All built the same way.
Last updated: July 2026. This is one non-developer's workflow, not a definitive guide. Tools change fast; what worked this month might be different next quarter. The pattern — spec, build, ship, iterate — holds up.
