How we rebuilt a Shopify store from a template into a custom theme with Claude Code

We rebuilt Golf Subculture, a print-on-demand golf cap brand, from a standard Shopify template into a custom theme using Claude and Claude Code. The headline change is a product-card system that renders every cap on the homepage as a WhatsApp-style group chat, a mate finding the cap and sharing it with the group. There was no developer, no theme from a marketplace, and the kind of bespoke storefront that normally needs a Shopify developer and a budget. You can see the brand live at golfsubculture.com.

This is a build case study, not a growth story. It is about what one person with commercial judgement and no coding background can now make, and it is honest about what that does and does not solve.

Starting from a template everyone else is using

The store launched on Shopify's standard Craft theme, which is a clean, capable starting point that has one problem: it looks like every other store on the standard Craft theme. For most brands that is a survivable issue. For this one it was not. Golf Subculture sells under the line Country Club Wear For Degenerates, aimed at golfers who reject the polished country-club aesthetic. A store that looked like a default template undercut the entire pitch in the first second a visitor landed. The brand needed a storefront that looked as irreverent as the product.

The usual route from there is to hire a Shopify developer, write a brief, wait for a quote, wait for a build, and pay an invoice that a pre-revenue brand cannot really justify. The route we took was to direct Claude Code to build the custom theme instead, working in the store's own Liquid templates, with me describing the outcome and reviewing what came back.

The build was directed, not hand-coded

The custom work is real front-end development, produced by describing what I wanted rather than typing it myself. The centrepiece is a reusable product-card snippet that Claude Code built and wired into both homepage carousels and the collection page. Each card presents as an authentic group chat. A friend discovers the cap and shares it with the group, the product photo sits in the middle of the conversation as an image message, and the add-to-cart and buy buttons are styled as chat bubbles in the thread. That is not a setting hidden in a theme somewhere. It is bespoke layout, markup, and styling, the kind of thing that only exists if someone builds it. I was the someone, in the sense that I decided what it should be and judged whether each version was right. Claude Code was the someone in the sense that it wrote the Liquid and the CSS that made it real.

The same approach built the rest. A hero chat box on the homepage carrying the brand's opening joke. A frequently-asked-questions page with a proper accordion that opens one item at a time. A contact page and an about page in the brand's look. None of it came from a template library. All of it came from describing the result and correcting the output until it matched.

What the iteration actually looked like

Directing the build was a loop, not a single instruction, and the loop is where the judgement lives. I would describe the group-chat card in plain language: a friend shares the cap, the product photo sits in the middle as a sent image, the buttons read as chat bubbles. Claude Code would build a version, and the first version was usually close and wrong in a specific way. The bubbles sat on the wrong side, or the image broke the conversation rather than sitting inside it, or the spacing felt like a form rather than a chat. I would say what was off, in the same plain language, and the next version fixed it. A few rounds of that and the card read like a real group chat instead of a developer's idea of one. The work I did was not writing code. It was knowing, as the person who understands the joke, when a version was funny and believable and when it was not, and being specific about the gap.

That same loop built the unglamorous structure around the showpiece. The nineteen products were organised into the right collection, the main menu and footer were rebuilt with proper categories, and a footer link that had been quietly returning a 404 was fixed. None of that is interesting to look at, and all of it is the difference between a store that works and a demo that photographs well. Doing it through Claude Code meant the boring structural work and the creative showpiece were the same kind of task: describe the outcome, review the result, correct it. It also meant I never lost the thread of what the brand was, because I was in every decision rather than handing a brief to someone and hoping the result came back in the right spirit. The store came together as one coherent build instead of a clever card bolted onto a template.

Why the whole thing runs on metafields

The part that matters most for a non-technical owner is that the custom system stays editable without touching code, because the content lives in metafields. Thirteen metafields hold the chat messages, which side of the conversation each speaker sits on, where the product image sits in the thread, and the text on the buttons, and they are filled in for all nineteen products. So when I want to change the joke on a particular cap, I open that product in the Shopify admin and edit a field, the same way I would edit a price. I am not editing code. The hard part, the system that turns those fields into a rendered group chat, was built once. The everyday part, writing a different conversation for each cap, is a form I fill in.

That separation is the whole trick, and it is the same principle behind every good build. Put the complexity in the system, where it is built carefully once, and leave the owner a simple, safe way to change the content as often as they like. A custom theme that only a developer can edit becomes a cost centre the moment you want to change a word. A custom theme driven by metafields stays yours.

The unglamorous fixes mattered as much as the showpiece

A redesign is not only the part you show people, and the dull fixes decide whether the showpiece is ever seen. Alongside the chat cards, Claude Code sorted a currency misconfiguration so the store now shows prices in US dollars to everyone, rebuilt the FAQ so its content is driven by a metafield rather than hardcoded, and worked the product-page performance score from 47 up to 57 through image handling and removing legacy CSS. A slow store with the wrong currency loses the sale before the visitor ever notices the clever product card. Spending the AI build time on the boring layer as well as the showpiece is what makes the showpiece worth having.

What an AI build changes about cost and speed

The commercial change is that a custom storefront stopped being a budget decision and became a time decision. The work that used to be a developer engagement, with a brief, a quote, a wait, and an invoice, became a run of working sessions where I described what I wanted and reviewed what was produced. For a brand that is still proving itself and cannot responsibly spend on a developer, that is the difference between having a distinctive store and settling for the template. It also keeps the brand owner in the chair. I made every design and copy call, because I am the one who knows what degenerate golf culture finds funny, and the build kept pace with those calls instead of routing them through someone who does not share the context.

The honest limit: a good build does not create demand

The redesign made the store distinctive. It did not make anyone buy, and pretending otherwise would be exactly the kind of thing this brand exists to take the mickey out of. Golf Subculture is live and still working the harder problem, which is demand and trust with an audience that has never heard of it. A stranger landing on a sharp, funny, custom store still has to decide that an unknown brand is worth the risk, and a beautiful storefront removes one objection without creating any of the want. That is the honest shape of it. The theme was the solvable problem, and solving it with Claude Code was fast, cheap, and genuinely good. The business question underneath, whether enough of the right people want these caps, is not a design problem and a redesign was never going to answer it. Knowing which of your problems a build can solve, and which it cannot, is most of commercial judgement. It is the thread through every business case study worth reading: what the work actually changed, and what it did not.

Get the free Ecommerce Roadmap at ortopylot.com. All 73 subjects, why each matters, and what happens to businesses that skip them.