We Built 14 Commerce Stores With AI. Ecommerce Templates Now Look Like a Weird Historical Accident

I work in marketing, so I am professionally trained to be suspicious whenever developers tell me something is “really cool.”

Usually, this means somebody has shaved 18 milliseconds off an API response or discovered a JavaScript framework that will be abandoned before I finish writing the campaign.

SUPERFAST SHOP is different.

I have watched it grow from a collection of experiments into something I genuinely think changes how we should demonstrate a commerce platform and build POCs, MVPs, and real stores with full industry-related features.

Today, 14 stores are live. A kitchen retailer. A camera shop. A laboratory supplier. A fashion store. A restaurant group. A streaming service. An auto-parts business. A food wholesaler. A tool supplier. Hotels. Furniture. Drinks distribution. A web shop inside the game and toy blocks builder.

They don't share a template because that would miss the point. Each of these demos starts with a different business model and the same question:

What would a commerce system look like if it were built around the business's needs, rather than forcing the business to fit an off-the-shelf e-commerce platform?

The answer is SUPERFAST SHOP: a catalog of production-shaped commerce systems built with AI, Crystallize, and the Crystallize MCP.

“Production-shaped” is important because these ARE NOT three fake products, a hero image, and a suspiciously functional Add to Cart button. Across the demos, you find bookings, rentals, subscriptions, approval workflows, contract prices, 3D product configuration, vehicle fitment, customer-specific rankings, in-game micro payments, regulated product data, multi-market pricing, hyper personalization, thousands of variants, 38 capabilities tagged across the stores.

With all of this☝️the usual “Does your platform support feature X?” conversation feels slightly primitive and outdated.

Let's start with the ridiculous stuff.

Kitchen e-commerce has a fairly obvious problem: the thing being bought does not really exist yet.

A kitchen is a system of cabinets, fronts, worktops, appliances, handles, dimensions, layouts, and questionable domestic negotiations about whether an island is really necessary.

Selling that with a standard SKU page is optimistic.

Kitchen Universe approaches the problem differently. The 3D configurator is the commerce experience.

You can start with one of six kitchens, move components around in 3D, switch finishes and worktops, and see pricing change based on the actual parts being used. The same product structure powers different markets and currencies.

Then it gets considerably more interesting. An AI concierge can construct the kitchen from a conversation. Ask for a long kitchen, a central island, wood finishes, and a specific appliance setup, and the agent works with the actual catalog and prices instead of hallucinating an inspirational kitchen from Pinterest and wishing you luck.

The catalog also generates print-ready PDFs from live product data and current prices. That means the website and the annual catalog don't need to become two estranged databases that occasionally meet at Christmas🎄

The business point is bigger than 3D.

When the product is configurable, the commerce architecture needs to understand the configuration as the product. Crystallize supplies the product model and commerce primitives; the MCP gives the agent a structured way to work with them.

The storefront then becomes a consequence of the business model rather than the thing dictating it.

Highlights/features:

• 3D product configurator
• An AI concierge that builds the kitchen
• PDF catalog from live prices
• Priced from parts per market
• ACP and UCP checkout for AI agents, with product feeds

Go ahead and build your dream kitchen > Kitchen Universe

Furniture has a similar problem to kitchens, with the added joy of inspiration content. Customers don't merely buy “sofa SKU 417.”

They see a room. They like a sofa. They want another module, a different fabric, maybe a shelving system beside it, and then someone has to calculate what this beautiful idea actually costs.

Furniture Universe connects those things.

This one will blow you away. Product configurators normally end with something resembling a product. You choose the color, add a component, or pick an option. Eventually, the system produces a configuration and gives you a price.

Block Builder Universe takes that logic to a slightly ridiculous (and genuinely useful) extreme.

Here, the build itself is the basket.

The builder has 1,056 blocks across 62 shapes and 20 colors. Every block you place in the 3D model corresponds to a real product variant in Crystallize. Add another brick, and you have changed the order. Remove ten, and the basket changes with it. A ready-made Castle contains 1,696 blocks, and the cart knows exactly which ones are required and what the complete build costs.

OK. Had fun? What did you make? Please stop burning our tokens.

Oh, that’s not all, folks. We take it up a notch when AI enters the room and refuses to behave modestly.

Instead of building everything manually, you can describe what you want (or provide an image), and the AI creates a parts plan using actual blocks available in the catalog. A prompt can become a 1,290-block bonsai tree, Booba Fett Bounty Hunter, Viking boat, or plumber Mario, which you can then inspect, modify, share, and order as the exact collection of parts needed to build it.

Shared creations become commerce too. Builds saved to the gallery can be reopened in the configurator, modified, and ordered. The system can also generate building instructions as a PDF.

This demo demonstrates a commerce model where the customer, configurator, or AI can effectively create a sellable product at runtime from real inventory.

You don't need a preconfigured SKU for every possible model anyone might invent. Crystallize holds the structured components and variants; the experience assembles them into something unique and keeps commerce logic attached throughout.

That pattern travels well beyond toys.

Furniture systems (as we showed). Industrial assemblies. Custom kits. Electronic components. Spare-part packages. Made-to-order products.

Highlights/features:

• 3D Product Configurator
• An AI that builds from a sentence
• Building instructions as a PDF
• One product, thousands of SKUs. 

3D Brick Builder with AI Model Generator for a unique experience > Block Builder Universe

Marketing has hurt personalization terribly. For years, changing a homepage banner from “Hello everybody” to “Hello Nebojsa” apparently qualified as a personalized digital experience. Recommendations are there as well, but NOT really “personalization”, right?

Photo Store sets the bar somewhat higher.

It takes customer order history and converts it into taste vectors. Those vectors influence campaigns, product grids, category pages, search, the mega menu, and even recommendations around the basket.

The result is effectively a different store for different customers.

A professional video shooter and a wildlife photographer can browse the same catalog and see the same campaigns ranked differently because their buying histories tell the system they care about very different things.

More importantly, the ranking is explainable. Product cards can show why something appeared where it did, and the demo includes tools that let merchandisers understand and work with the ranking rather than surrendering the storefront to the sacred mystery box known as “the algorithm.”

That's the business value of hyper-personalization and Rank by Taste: not “AI recommendations,” but using the same catalog more intelligently for every customer.

No cloned catalog. No manually maintained customer segments for every possible taste. One product universe, ranked differently depending on who walks through the door…I mean home page.

Highlights/features:

• Vector ranking everywhere (hyper-personalization)
• Explainable product ranking
• Merchandiser vector lab
• Typo-tolerant site search

Deliver a different shopping experience for every shopper > Photo Store Universe

Hotel/apartment booking websites have spent years teaching us that booking needs to live in a separate technological kingdom.

Product catalog over here. Booking engine over there. Experiences in another system. Availability somewhere deep underground, guarded by an enterprise integration consultant.

Boutique Universe asks a much simpler question: what if time and availability were just part of the commerce model? What if time becomes inventory?

The demo contains 40 hotels, villas, and unusual stays in Norway, with rooms drawn from available pools and experiences sold in time slots. Availability, holds, and confirmation run through the Crystallize Shop API.

Then an AI concierge sits on top of that structure. It can search the catalog, understand the customer's preferences, check availability, and hold dates. Discovery can also rank destinations and stays by the guest's taste and explain why it recommended each option. And suddenly, finding a more remote, rustic, and cold place becomes easy (not that it is hard to find one in Norway🤔).

That's interesting for hotels and apartment rentals, obviously. But the bigger idea applies to anyone selling finite access to something: activities, classes, equipment, venues, consultations, TOOLS, or experiences.

Once booking becomes a native commerce primitive, you stop building “e-commerce plus booking system.”

You just build commerce.

Highlights/features:

• Online booking in e-commerce
• Experiences in time slots
• Taste ranking with reasons
• AI concierge

Enhance your Hotel Booking with > Boutique Universe

BTW, I mentioned tools. Battery platform matters. Technical specifications matter. Depot inventory matters. Sometimes you buy the drill. Sometimes you rent the concrete saw. Sometimes you need a quote.

Have a look at Tools Universe; it contains 416 products from six manufacturers with typed, ETIM-aligned specifications. You can buy hammers, drills, batteries, and parts. Heavy equipment, however, works differently: 24 rental items are available across five depots and can be booked by the day/week, with services like site delivery and damage waiver attached. Trade customers can also request quotes.

This demo combines multiple commerce models without turning them into separate websites.

Buy. Rent. Quote. Filter on actual technical characteristics. Check local stock. One commerce system. Bookable commerce at its best.

Restaurants are a good example of why “e-commerce” is sometimes the wrong mental model. A restaurant doesn't necessarily want to ship anything. It wants to fill tables.

Wine and Dine Universe represents five restaurants from one setup. Guests choose a restaurant, date, seating, and party size, then book an actual table on a floor plan.

Each table effectively becomes a bookable unit with availability.

Events use the same underlying idea. A wine tasting can be booked against tables and seats rather than being shoved into an external ticketing application because “our CMS doesn't do that.”

Another detail I particularly like is that you can sync drinks from Spirits Universe, the supplier demo, while the restaurant controls its own selling prices. Individual dishes can then recommend wines or alcohol-free pairings from that supplier catalog.

Now we're no longer showing isolated storefronts. We're showing commerce systems talking to commerce systems.

A supplier maintains its product universe. A restaurant consumes relevant parts of it. Both businesses retain their own commercial logic.

Which starts looking a lot more like the real economy than another demo selling three T-shirts.

Highlights/features:

• Native booking with time-slot bookings
• Tables on a floor plan
• Events booked per table
• Drinks synced from Spirits Store
• Estimated spend per booking
• Subscriptions for Spirits

Enhance your Restaurant Table Booking and Event Reservations with > Wine and Dine Universe

I’ve mentioned Spirits Universe earlier; what’s up with it? Well, it takes another route into B2B complexity.

A restaurant, bar, shop, and importer may technically buy bottles from the same supplier, but commercially they are not the same customer. This is where “customer-specific pricing” stops being a checkbox on a feature matrix.

Price is only one expression of the customer relationship.

The real opportunity is a storefront that understands who is buying, under what agreement, in which market, with which permissions, and for what commercial purpose.

Once that context becomes structured, a B2B store starts behaving less like a public catalog with a login screen and more like an actual sales channel.

Most e-commerce businesses start with a storefront.

Troll Universe starts with a forest, a troll, and a merchant standing beside a cart.

We made a 3D browser game, Troll, where commerce isn't a separate shop bolted onto the experience. It is part of the world. Players explore five environments inspired by Norwegian nature and folklore, meet merchants, and buy cosmetics, equipment, bundles, and game access using real Crystallize products, carts, and orders. Four merchants offer 43 items for sale, with product names, images, and prices sourced from Crystallize.

This isn't merely a themed product catalog.

A cap can have three, five, or 10 color variants, each with its own image and price, and players can try it on before buying. Deals and bundles are represented as price variants. Players can buy items for themselves or gift them to another player. Digital goods can even follow different ownership models: rent something temporarily or buy it once and keep it.

Gaming commerce often lives in a parallel universe of custom inventory systems, virtual currencies, and bespoke store logic. Troll Universe offers another option: digital items can behave like structured commerce products, with the buying experience embedded directly in whatever digital world the customer is exploring.

The shop doesn't need to look like a shop.

It needs products, prices, variants, customers, and transactions. What you put around those primitives can be a traditional storefront or a troll wandering through a Norwegian forest looking for a better sweater.

Highlights/features:

• A web shop inside the game, Micore payments
• Deals and bundles as price variants
• Try on before you buy
• Gifts to other players

In-Game Store and digital goods selling with > Troll Universe

Fashion ecommerce excels at generating variants. Unfortunately, we humans don't wake up desperately wanting “variant ID 812991-BLU-M.”

Fashion Universe models ten brands as different style tribes and uses shopper taste to re-rank the experience. It combines that personalization with 64 shoppable looks, where editorial photography connects directly to sellable colorways and sizes.

Behind the visual layer lies much more serious product data: the demo includes 12,575 color and size variants, with composition, supply-chain information, and CO₂e data attached to each garment.

This example shows why structured product data matters more than any architecture diagram ever will. Good product modeling doesn't just make the PIM cleaner.

It creates things you can sell with.

The same underlying data powers variants, filtering, localization, storytelling, campaigns, personalization, and transactional content.

That is a very different proposition from “our API returns products.”

Highlights/features:

• Personalized by style (Personalized e-commerce with vector ranking)
• Shoppable looks
• Colorways and sizes (One product, thousands of SKUs)
• Supply chain per garment

Your fashion brand website can look like this > Fashion Universe

Selling car parts online presents a wonderfully brutal usability requirement.

The customer does not care that you have 429 parts. They care whether one of them fits their Porsche 911 Turbo (930) from 1976 currently occupying their garage😎

Car Parts Universe therefore starts with the vehicle. Enter a registration number and the storefront narrows itself to compatible parts. Users can search by OE number, check availability at their branch, and see when a part will arrive in the next delivery round.

Workshop customers then get the B2B layer: their own trade pricing, invoices, quotes, and fleet information.

The important capability here is vehicle fitment, but the business lesson is broader. Sometimes search shouldn't start with the product. It should begin with the customer's context.

Car. Machine. Location. Existing equipment. Previous purchase. Contract. Compatibility requirement.

Once your commerce backend can represent those relationships properly, discovery becomes dramatically more useful.

“Search 429 products” becomes “show me the parts that solve this specific job.”

That's a fairly meaningful upgrade.

Highlights/features:

• Vehicle fitment
• Typo-tolerant (OE number) search
• Stock per location
• Vector personalization
• Invoice and credit limits

Build your store to deliver parts customers actually need > Car Parts Universe

B2B food e-commerce sounds deceptively straightforward until you meet a professional kitchen.

Pack sizes matter. VAT matters. Allergens really matter. Case pricing matters. Credit matters. And chefs tend to think in dishes, not database rows.

Food Universe combines those requirements in a foodservice storefront built around GS1 product data and trade purchasing.

It contains 435 products from 63 brands, with pack information and allergens available as structured data and filters. Business customers order on invoice within credit limits and can buy using volume tiers.

Then content enters the transaction.

Recipes aren't decorative SEO pages floating around the side of the store. They understand their ingredients, so a kitchen can move from recipe to purchasing those ingredients.

And because subtlety was apparently abandoned during this project, we built the demo in less than a day.

The interesting part isn't “AI builds websites fast.” AI can generate websites. We know.

The interesting part is that an agent working through the Crystallize MCP can build against structured commerce primitives quickly enough that the conversation moves from generating HTML to modeling an actual business.

That's much more useful.

Highlights/features:

• Shoppable content (recipes fill the basket)
• Volume and tiered pricing
• Invoice payment and credit limits
• Detailed product data (GS1 and allergens)

Sell products and recipes and allow customers to choose ingredients > Food Universe

This one might be the best example of what “production-shaped” actually means.

Because Lab Universe is not sexy e-commerce. It is procurement. And procurement is where innocent Add to Cart buttons go to discover pain.

Different buyers have different roles. Departments have approval thresholds. Cost centers have budgets. Institutions negotiate contract prices. Some items repeat on standing orders. Regulated chemicals need hazard information and additional controls.

Lab Universe models those requirements directly.

The store contains 2,211 SKUs across seven manufacturers. Institutions see their own contract prices. Departments have approval logic. Buyers work within cost centers and budgets. Product families have typed specifications, and regulated chemicals carry GHS hazard data.

This is B2B ecommerce after somebody has finally spoken to a B2B buyer. The point isn't that Crystallize has an “approval workflow feature.”

The point is that you can model the organization, product data, commercial agreements, and purchasing process as one coherent commerce system.

That's the difference between putting a B2C checkout in front of a company and actually doing B2B commerce.

Highlights/features:

• Subscriptions and approval flows
• Budgets and customer price lists
• Volume tiers
• Buyer roles

Build yourself an easy lab supplies B2B eProcurement > Lab Universe

Subscriptions suffer from another software-induced assumption: people hear “subscription commerce” and immediately picture coffee bags arriving every four weeks.

Useful. But rather limiting.

Screen Universe uses Crystallize subscriptions to model a streaming service.

There are free trials, monthly and yearly plans, upgrades, pauses, and metered downloads. Films can also be rented for 48 hours or purchased outright, meaning recurring and one-off transactions coexist rather than being separated into competing payment realities.

It also applies personalization per household profile. Change the viewer, and the home page, and content rows and recommendations change with them.

Again, the business model comes first.

A “product” might be a physical item. Or an entitlement. Or access for a month. Or 48 hours of viewing. Or usage measured against a plan.

Commerce gets much more interesting once you stop defining it by the shape of a shipping box.

Individually, each demo proves something useful. Together, they make a much stronger argument.

A template starts by asking: Which parts of your business can fit into the structure we already designed?

SUPERFAST SHOPS starts with the opposite question: How does your business actually work?

Do you sell things, access, time, or configurations? B2C, B2B, or both?

Do customers negotiate prices? Need approval? Rent equipment? Configure products? Book resources? Buy subscriptions? Need recommendations based on taste? Need products filtered by the machine, vehicle, or system they already own?

Great. Now, model that. Then build the commerce experience around it.

This is also why AI changes the equation more than I/we all initially expected.

AI-generated storefronts by themselves aren't particularly interesting anymore. Give a capable model enough prompting and caffeine-equivalent compute, and it can produce something resembling an e-commerce frontend.

The bigger/harder problem is everything behind it.

• What is the product model?
• What relationships matter?
• How should prices behave?
• What does availability mean?
• Who can buy?
• What can they see?
• What needs approval?
• Which specifications drive discovery?
• What context should an agent understand?

That is where Crystallize's amazing content modeling abilities and our MCP come in. Where SUPERFAST examples come in. The agent isn't starting from a blank code editor and inventing a commerce universe. It can work against structured products, content, customers, markets, orders, bookings, subscriptions, and other commerce primitives.

SUPERFAST SHOP is what happens when you give it something meaningful to build with.

At the bottom of the SUPERFAST SHOP website, there's a Build yours with AI flow. It asks three things:

• What kind of store are you starting from?
• Who do you sell to?
• Which capabilities do you need?

Choose the relevant pieces from the demos, and it will generate a prompt to build your version in Crystallize.

You don't have to start with one demo either. That's the point.

Take the product configurator from Kitchen Universe.

The B2B pricing logic from Spirits Universe.

The procurement workflow from Lab Universe.

The taste ranking from Photo Store.

The booking model from Boutique Universe.

Congratulations. You've just described a business that would make most e-commerce template marketplaces quietly close the browser tab.

The generated build also assumes some things should simply be standard in 2026: server rendering, canonical URLs, hreflang and sitemaps for SEO; JSON-LD structured data; llms.txt, llms-full.txt and Markdown twins for AI discovery; a clean accessibility tree and WebMCP actions for agents; and performance targeting Lighthouse scores near 100.

Which brings me back to why this whole thing impressed me in the first place.

SUPERFAST SHOP is not really about demo stores. It shows that modern commerce doesn't need to start with a generic storefront and work backward.

Start with the business. Model how it actually works and let AI help build the system around it.

Then make it superfast.

Build yours with AI on Superfast.shop.

SCHEDULE A 1-on-1 DEMO to explore what is possible today with Crystallize. Or, why not, sign up and  START building for FREE.