Core Thesis

  • For three decades, digital retailers have asked shoppers to navigate fixed websites that reproduce the logic of a physical store.
  • While enterprise AI has yet to pay for itself, retail leaders are accelerating investment. The long-term vision is that AI will unlock control over how, and whether, a customer ever encounters a storefront at all.
  • Two architectures are emerging in response:
    1. Generative storefronts that assemble themselves around each visitor's purpose.
    2. Autonomous agents that complete routine purchases without a storefront appearing at all.

Since its inception, online retail has been a remarkably literal transcription of the physical store.

Diagram mapping physical-store elements to their online-store equivalents: front door to homepage, aisles to categories, shelves to product pages, register to checkout.

A Reckoning, and a Doubling Down

Since AI entered the corporate mainstream, a broad consensus has formed that it will remake the customer journey in retail as fundamentally as e-commerce once remade the physical store. The evidence for that claim, however, has grown more equivocal even as the claim itself has grown louder.

McKinsey’s State of AI in 2026 survey found that despite record enterprise investment, 63% of organizations report no EBIT impact from AI use, and 1/5 cite AI-related operating costs as an active constraint on further deployment.1

Which begs the question: If the return on enterprise AI remains this unproven, why have online retailers accelerated their commitments, in some cases by an order of magnitude?

The investment record offers a partial answer. Amazon’s Rufus shopping assistant, built on custom silicon and trained on the company’s own product catalog, already answers tens of millions of customer queries a day. Other retailers have their own AI chatbots: Walmart has Sparky and Wayfair has Muse.2

The retailers moving fastest are not simply layering a chatbot onto an existing website but are also beginning to treat the storefront as something the system can construct in real time. In some cases, retailers view the storefront as an intermediary that AI can bypass altogether.

The AI Era of Online Retail Will Arrive in Three Stages

Diagram of the three stages AI will reshape online retail. Stage 1: AI sits inside the storefront (conversational search, styling, recommendation). Stage 2: AI becomes the storefront (the retailer dynamically composes products, content, navigation and presentation around the shopper's mission). Stage 3: AI becomes a persistent commerce layer (the system remembers preferences across sessions and channels, and determines when a routine purchase should be automated).

As the stages unfold, retailers will retailers and brands will separate experiential commerce from utility commerce with much greater precision. A shopper assembling a wardrobe or furnishing a home may still want a rich branded environment, while one reordering detergent will increasingly prefer the transaction to disappear into automation entirely.

Two Futures for Online Retail

1. Custom Online Storefronts Built by AI

Brunello Cucinelli, the Italian fashion house, spent three years building Callimacus with Salesforce CEO Marc Benioff.

Callimacus describes a “new web” without pages, categories, or fixed paths, allowing the interface to organize itself around a visitor’s purpose. Using approved product, editorial, and design components, the system infers intent and assembles the experience in real time.

What looks like an idiosyncratic luxury experiment may prove to be the first credible prototype of the post-website store.3

The generative storefronts Callimacus creates can help construct preferences through exposure, comparison, and dialogue, leading to a more tailored, luxurious online retail experience, as if you were shopping with the aid of Cucinelli’s elite store staff. In turn, this unique customer journey should lead to a more frictionless shopping experience, with greater conversion, higher average order values, and more durable customer loyalty.

Luxury Crosses the Economic Threshold First

It is no surprise that Cucinelli got here before H&M, the Swedish fast-fashion giant. On each company’s most recently reported full-year results, Cucinelli’s gross margin was 75% against H&M’s 53%. Applied to a representative transaction in each company’s category, that gap works out to roughly $233 of Gross Profit for Cucinelli versus $43 for H&M.4

Luxury also has more strategic value at risk because Cucinelli sells symbolic meaning, connoisseurship, and a particular conception of taste alongside the garment itself. When an external agent reduces a cashmere sweater to a thumbnail, price, and specification beside cheaper alternatives, the brand can lose the contextual cues that sustain willingness to pay. By supporting conversion while protecting brand meaning, Callimacus addresses both commercial and strategic objectives.

The early evidence warrants cautious optimism, with Callimacus reporting that active time roughly doubles after visitors begin interacting with the system, although no controlled estimate of incremental profit has been published.5 Large randomized retail experiments have also found sales gains as high as 16.3% from selected generative-AI workflows, driven mainly by conversion and varying materially by use case.6

Falling Deployment Costs Will Broaden the Market

Chart showing the all-in cost of a personalized AI session falling over time, crossing the viability threshold first for luxury / high-customer-value retail, then broader high-consideration categories, then mass / low-margin retail.

Current economics offer a poor guide to the long-run market. The price of reaching GPT-4 model performance has fallen from roughly $30 per million tokens in March 2023 to about $0.10 per million tokens for comparable open-weight models by mid-2026, a roughly 300x decline in three years.7 Token prices capture only part of the cost curve, since smaller models, intelligent routing, caching, and standardized integrations also reduce the expense of a personalized session.

Chart showing the lowest publicly available price to reach GPT-4-launch-level performance falling from $30.00 per million tokens in March 2023 to $0.10 per million tokens by mid-2026, log scale — more than 300x cheaper.

Cucinelli spent more than three years building Callimacus. The platform now integrates with Salesforce and Shopify, and Salesforce’s investment this summer is intended to convert proprietary development into repeatable enterprise infrastructure. As both intelligence and implementation become cheaper, the adoption threshold should move from luxury to high-consideration categories and eventually toward mass retail.8

2. Agentic AI Purchases

A second architecture bypasses the storefront altogether, and Shopify has bet directly on it.

Shopify’s CEO, Tobi Lütke, has spent much of the past two years reorganizing the company’s product roadmap around what he calls agentic commerce, the expectation that AI assistants will complete purchases outright. The centerpiece of that bet is Shopify Catalog.

Shopify Catalog is a database indexing billions of products across Shopify’s roughly two million merchants, organized by the structured attributes (e.g., price, availability, color, fit) that can be parsed and acted on directly.9

In June 2026, Shopify and Google jointly enabled the Universal Commerce Protocol (UCP) by default on every Shopify store. The UCP is the pipeline that lets an AI agent discover a merchant’s checkout, catalog, and fulfillment capabilities and complete a transaction without a human ever loading a page. Shopify paired the protocol with Agentic Storefronts, letting any merchant surface its products and brand voice inside ChatGPT, Gemini, and Microsoft Copilot.10

Sam Altman has endorsed the bet from the outside. In an August 2026 interview, the OpenAI CEO said of Lütke’s investments in AI: “He is always six, eight months ahead of any other CEO.”11 Altman has also invested in this future of retail through OpenAI’s own Instant Checkout and Agentic Commerce Protocol with Stripe, simultaneously building a competing rail for the very same transactions.

McKinsey’s Agentic Commerce report states that agentic commerce along these lines could generate as much as $1 trillion in U.S. B2C retail revenue by 2030, and $3-5 trillion globally.12

In this third stage, AI agents, not the shopper, increasingly decide whether a branded storefront is worth rendering at all. For a retailer, the operative question is no longer only how good its website is, but whether its catalog is legible to an agent it does not control.

What This Means for the Online Retail Landscape

The honest answer to whether one of these architectures produces a single winner is: it depends on which layer of the stack is in question.

The protocol layer (the rails that let an agent discover a catalog, negotiate a price, and complete a checkout) behaves like infrastructure. And infrastructure tends to consolidate. This infrastructure will benefit from the same network effects that produced a small number of winners in cloud computing, mobile app stores, and card payments. Shopify and Google’s UCP, OpenAI and Stripe’s ACP, and Amazon’s own rails are early bids for that position.

The experience layer sits on top of that infrastructure and behaves differently. A Cucinelli garment is not made more valuable because more shoppers use Callimacus. A brand’s design grammar, product knowledge, and earned trust are not subject to the same network. The closer analogy is the relationship between electricity and the appliances plugged into it: a small number of standards can underpin an unlimited number of genuinely differentiated destinations.

Footnotes

  1. McKinsey & Company, “The State of AI in 2026: On the Road to ROI,” August 2026.
  2. Amazon.com Inc. FY2026 capital expenditure guidance, as reported in 24/7 Wall St., “Amazon’s AI Investments Are Creating a Whole New Business Model,” August 28, 2026; 24/7 Wall St., “This Retailer Might Rise as a Top Stealth AI Play,” May 19, 2026; Ralph Lauren Corporation, “Ralph Lauren Introduces Ask Ralph, a Conversational AI-Powered Styling Experience,” September 9, 2025; Zalando, “Zalando Brings Its AI-Powered Assistant to All Markets,” October 2024; Wayfair, “Wayfair Introduces New AI-Powered Tool ‘Muse’ to Inspire and Personalize the Home Shopping Experience,” February 11, 2025.
  3. Brunello Cucinelli S.p.A., “Analyst Presentation: First Half 2026 Results,” July 30, 2026, 12–13; Callimacus, “The Platform That Composes Your Website Around Whoever Arrives,” accessed August 11, 2026.
  4. Estimate. Gross margins are actual, most recently reported full-year figures: Brunello Cucinelli S.p.A., 75.2 percent (WWD, “Brunello Cucinelli Talks Saks Global, Reports Strong 2025 Growth,” February 18, 2026); H & M Hennes & Mauritz AB, 53.4 percent (“Full-Year Report 2025,” January 30, 2026). Average order values are category-level estimates, not company-disclosed figures — approximately $310 for luxury apparel and $80 for fast fashion (Statista, “Luxury Apparel Average Online Order Value,” 2025; Eightx, “Average AOV by Ecommerce Vertical 2026,” June 2026). Neither company discloses transaction-level economics, so treat the dollar figures as illustrative rather than reported.
  5. Zwieglinska, “Luxury Briefing.”
  6. Lu Fang, Zhe Yuan, Kaifu Zhang, Dante Donati, and Miklos Sarvary, “Generative AI and Sales Productivity: Field Experiments in Online Retail,” working paper, revised June 30, 2026.
  7. Value Add VC, “AI Inference Cost Reduction in 2026,” June 23, 2026, drawing on published pricing from OpenAI, Google, and DeepSeek and on a16z’s “LLMflation” analysis.
  8. Brunello Cucinelli S.p.A., “Analyst Presentation”; Callimacus, “The Platform”; Zwieglinska, “Luxury Briefing.”
  9. Fast Company, “Shopify Has Gone All In on the Agentic Commerce ‘Gold Rush,’” March 24, 2026; Dealroom.co, “Shopify Turns On UCP and the Catalog: Agentic Commerce for Every Store,” June 19, 2026.
  10. Shopify, “Introducing Shopify Agentic Storefronts: Sell Your Products Everywhere AI Conversations Happen,” December 10, 2025; PageFly, “Shopify Agentic Commerce: How to Sell in ChatGPT, Copilot & Perplexity,” June 9, 2026.
  11. David Senra, “Sam Altman, OpenAI,” Founders podcast, August 2026, https://youtu.be/kG8AoExkX40 (quote at 0:50).
  12. Katharina Schumacher, Roger Roberts, and Katharina Giebel, “The Agentic Commerce Opportunity: How AI Agents Are Ushering in a New Era for Consumers and Merchants,” McKinsey & Company, October 17, 2025.

Works Cited