The Storefront That Builds Itself

Every platform is adding AI to commerce right now.

AI product descriptions. AI-generated images. AI chatbots. AI recommendations. The features keep stacking up, each one promising some percentage improvement in conversion or efficiency.

But here’s what I keep wondering: Are we adding AI to the wrong layer?

The Template Problem

Think about how an e-commerce store actually gets built today:

1. Merchant picks a theme or template
2. CMS renders a static page structure
3. Products get slotted into predefined grids
4. “Personalization” means swapping content within fixed containers
5. Customer sees the same layout whether they’re hunting for a specific SKU or browsing for inspiration

The page is designed once. The customer adapts to it.

We’ve spent two decades optimizing what goes into the containers while treating the containers themselves as immutable. The sidebar is always there. The product grid is always 4 columns. The hero banner is always 16:9.

What if the containers are the problem?

Netflix Already Solved This

When you open Netflix, you don’t see the same interface as me. You see your interface. The categories are different. The row order is different. The artwork is algorithmically selected for you. The entire page is generated based on your viewing history, time of day, device, and hundreds of other signals.

Netflix doesn’t have “templates” in the way e-commerce thinks about them. They have a layout generation engine that outputs whatever structure is most likely to get you watching.

E-commerce never made that architectural leap.

We kept the template-first model from 2005 and layered personalization on top. We can change which products appear in the “Recommended for You” carousel, but we can’t decide whether to show a carousel at all. We can change the hero image, but we can’t decide whether a hero image makes sense for this particular user in this particular moment.

The structure is fixed. Only the content is variable.

What Generative Storefronts Would Mean

Here’s the mental model I’ve been building:

Inputs:

• Customer context (CDP signals, session behavior, purchase history, device, location, time)
• Product catalog (inventory, pricing, margins, trends, competitor signals)
• Content library (images, copy variants, layout components, promotional assets)
• Business rules (margin targets, inventory priorities, brand guidelines)

Output: Not a page with personalized content—but a generated page structure optimized for this customer in this moment.

For a Hunter who knows exactly what they want: Dense product grid. Quick-buy actions. Minimal distraction. Price comparison surfaced immediately.

For a Gatherer in discovery mode: Magazine-style editorial layouts. Hero sections with mood-setting imagery. Story-driven product presentations. Slower, more immersive.

Same catalog. Same brand. Completely different experience. Generated per-session, per-user, per-context.

Why This Is a Platform Play

Individual merchants can’t build this. The infrastructure requirements are massive:

• Layout generation models that understand commerce UX
• Real-time rendering pipelines that don’t kill performance
• Validation systems that prevent hallucinated layouts from breaking checkout
• A/B testing frameworks that work with infinite layout variations
• Analytics that can attribute conversions to layout structure, not just content

This is platform-level infrastructure. It needs to be built once and deployed across thousands of stores.

And here’s the strategic reality: The platform that cracks this creates a moat.

Once a merchant experiences “the store that understands my customers better than I do,” going back to static templates feels like going from Netflix to a DVD catalog menu.

The Mid-Market Window

The enterprise players (Salesforce, SAP, Adobe) are too slow and too invested in existing architecture to make this leap. Their customers need migration paths and backwards compatibility. Radical reinvention isn’t in their DNA.

The pure-play marketplaces (Amazon, eBay) have different incentives. They’re optimizing for their platform, not for merchant brand experiences.

The mid-market platforms—the ones that own the full stack but aren’t encumbered by legacy enterprise complexity—have a window.

They can:

• Ship architecture changes to their entire merchant base simultaneously
• Iterate quickly based on real production data
• Position “generative storefronts” as a competitive differentiator against both upmarket and downmarket alternatives
• Build the merchant relationship around outcomes, not features

The first mover here doesn’t just add a feature. They redefine what a commerce platform does.

What the System Needs

Building this isn’t trivial. But it’s not science fiction either. The components exist:

1. Context Ingestion Layer Unify customer signals from CDP, session behavior, and external sources (weather, events, trends) into a real-time context object.

2. Layout Generation Engine LLM-powered system that takes context + catalog + constraints and outputs a layout schema—not just “which products” but “what structure.”

3. Component Library Pre-built, validated UI components (hero sections, product grids, editorial blocks, quick-buy cards) that the generator can compose.

4. Validation Pipeline Ensure generated layouts are functional (links work, checkout accessible), brand-compliant (no rogue styles), and performant (no 10-second renders).

5. Learning Loop Attribute conversions back to layout decisions. Which structures work for which contexts? Feed that back into the generation model.

This is complex. But it’s the kind of complex that platforms are built to handle.

The Question I’m Sitting With

Is the market ready for this?

Merchants are conservative. They like control. The idea of “AI generates my store layout” might feel like losing the keys to your own brand.

But here’s the counter-argument: Merchants don’t manually set Netflix’s row order. They don’t manually adjust Spotify’s homepage. They trust the system because the system works.

The real question isn’t whether merchants will accept AI-generated layouts. It’s whether the outcomes are good enough that they want it.

If generative storefronts convert better, retain better, and require less manual merchandising effort—merchants will adopt.

The platform that proves those outcomes first wins the positioning.

Where I’ve Landed

E-commerce is stuck in a template-first paradigm that made sense in 2005 but doesn’t make sense now. The technology exists to generate storefronts in real-time based on customer context. The platform that makes that architecture real captures a market position that’s very hard to compete with.

This isn’t about adding AI features to existing templates. It’s about making the template itself intelligent.

That’s the leap I think commerce needs to make. And I think the window for being first is narrower than people realize.

---

Related: The Infinite Concierge and The Role That Doesn’t Have a Title