# Inside AI’s $5 trillion quest to develop taste > Source: > Published: 2026-06-26 11:00:00+00:00 The assignment is charming: I’ve been asked to moderate a panel in a garden in early summer. The problem is that it will require an outfit I don’t own and have no time to find. Easy enough, I think. I use [AI](https://www.fastcompany.com/section/artificial-intelligence) every day to summarize transcripts, synthesize financial data, and draft emails. Surely I can outsource this. I open ChatGPT and ask it to find me a dress from my favorite label, Sézane, that’s appropriate for an outdoor professional event. It pulls dresses from the French brand’s catalog that are made of linen and crocheted—floaty, unstructured things that require a level of undergarment coordination I don’t want to think about. I clarify. It pivots. After several rounds of negotiation, it finally surfaces a red polka-dot dress I like. There are no links. I go to Google, search manually, and discover that the dress is from last season and available only in a size 2 on Poshmark. Twenty minutes after starting my search with ChatGPT, I give up. I head over to Sézane’s website and buy something the old-fashioned way. Artificial intelligence has, in the span of a few years, displaced coders, passed the bar exam, and written term papers for an entire cohort of college students. Yet, the technology that’s remaking civilization still can’t help me buy a dress. The retailers who crack this code are likely to own the next era of online shopping. When ChatGPT launched in 2022, becoming the fastest-adopted consumer technology in history, it lit a fuse. Behemoths like Amazon and Walmart, along with smaller players, raced to roll out AI-powered assistants to help customers find the right product from their catalogs. Soon startups like Daydream, Phia, and Remark were pulling in millions in venture capital funding to build AI shopping agents that could offer hyperpersonalized product recommendations. Over the past year, OpenAI, Google, and Anthropic have been working behind the scenes with the e-commerce infrastructure giants Shopify and Stripe to build [essential retail plumbing](https://www.fastcompany.com/91503503/shopify-most-innovative-companies-2026), such as AI-native checkout systems and real-time inventory updates. They’re preparing for a world of [agentic commerce](https://www.fastcompany.com/91533534/shop-til-you-bot-google-openai-and-the-race-to-build-agentic-commerce), where autonomous agents browse, compare, and buy on your behalf. Very soon, these companies promise, your AI assistant will know you well enough to anticipate your weekly grocery run and order the gear you’ll need for an upcoming ski trip, without you lifting a finger. Already 2% of all ChatGPT queries—50 million a day—involve shopping. McKinsey estimated last October that AI assistants could enable up to $1 trillion in U.S. shopping annually by 2030, and up to [$5 trillion globally](https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-agentic-commerce-opportunity). But as with so much AI-enabled technology, there’s a chasm between promise and reality. Right now, the experience of AI-assisted shopping is frustratingly ineffective. Walmart’s Sparky, perhaps the most advanced shopping chatbot, is still only slightly better than a search bar when it comes to certain things. Most frontier models, meanwhile, can’t tell if a product is in stock, much less let you click to buy it. “The vast majority of consumer-facing assistive gen AI experiences are prematurely launching because everyone has FOMO,” says Emily Pfeiffer, a principal analyst at Forrester who covers commerce technology. “They’re poorly tested.” Industry experts—including those leading agentic commerce efforts at Google and OpenAI—emphasize that change is coming. By the end of this year, they say, AI systems will be able to recommend products and automate checkout with far more ease than I experienced. But once these issues are resolved, there remain deeper, more philosophical hurdles to contend with. To recommend certain products, AI needs to develop an aesthetic sensibility. Consider my search for a dress. I told ChatGPT that Sézane is my favorite label, but even now I struggle to explain why. I like the way its silhouettes and palettes allow me to embody an idea I have of a French lifestyle; it conjures a woman who is unhurried, effortlessly stylish, and likely to be spotted at a farmers market. To serve my needs, the AI will need to understand my personal style, my emotional desires, and my evolving relationships with brands. Right now, no agent is anywhere close to achieving this. Can we teach machines to understand the complexities of human taste? Can they learn the nuances of a brand—not its logo and price points, but the feeling it produces in a person? It’s an ambitious task, but that’s exactly what technologists are working on now. And retailers everywhere have a lot riding on the outcome. AI is definitely helpful for specific kinds of shopping. When my washing machine finally conked out, instead of spending hours toggling between tabs to compare specs and reviews, I gave Google’s Gemini chatbot the dimensions of my punishingly small laundry closet and asked it to find me a new machine. It pointed me toward a GE model, which I bought at Home Depot. It’s been working like a dream. There’s a consensus that, in the near term, the best uses for AI shopping will be unglamorous. According to Neel Ajjarapu, product lead for commerce at OpenAI, ChatGPT is best at helping users shop for spec-heavy products that require a lot of research, like electronics, appliances, and sporting goods. Walmart’s Sparky, meanwhile, is great at helping you find tires. “Nobody knows what kind of tire they have,” says Daniel Danker, who leads AI acceleration at Walmart. “But if you tell Sparky you have a 2021 Rav 4, it will just ask you how many tires you need and [tell you] which one of our auto care centers has them in stock.” Yet actually purchasing these tires, light bulbs, and tennis rackets through a third-party chatbot is another matter. Large language models (LLMs) are trained by scraping the written output of the internet, and while they can see a retailer’s product pages, they can’t access crucial back-end information like inventory availability, or handle payment processing or customer support. That’s why OpenAI and Stripe developed the Agentic Commerce Protocol in January to create a shared language between businesses and agents, and why Google and Shopify launched the Universal Commerce Protocol to handle more complex issues like linking loyalty accounts to checkout. Shopify’s Catalog, meanwhile, acts as a bridge, allowing an agent to see a merchant’s inventory and pricing and surface relevant products in an AI chat. “What matters most isn’t which protocol dominates,” says Emily Sands, head of data and AI at Stripe. “It’s that there are open standards, that there is interoperability, that the important players are collaborating.” Streamlining the checkout process has proven especially tricky. Last September, OpenAI announced it was collaborating with Shopify to launch Instant Checkout—which would let users click to purchase directly from ChatGPT—only to walk it back six months later due to technical hurdles and a lackluster user response. Google, however, successfully debuted direct checkout features in Gemini and AI mode in search this past January. “People think that checkout is simple, but there’s a matrix of possibilities to think about,” says Vidhya Srinivasan, Google’s VP and general manager of advertising and commerce. “There are coupon codes and loyalty programs, and sometimes shipping requires a signature, all of which vary based on product and brand.” But in the world of AI, things move quickly, and today’s problems will soon be fixed and replaced by new ones. “Every couple of months, we see such massive changes that it’s impossible to predict what’s going to happen on what timeline,” says OpenAI’s Ajjarapu. “We’re planning by the seat of our pants.” When it comes to the subtleties of brand affinity—how that Sézane dress makes me feel—shopping assistants struggle. In the realm of style and taste, AI has to account for far more than a product’s price and materials. It needs to understand what its brand symbolizes to consumers. And that means that to survive in an era of AI shopping, brands must not only impress human beings; they need to be legible to machines. Traditionally, brands have tried to capture consumers’ attention visually, through billboards, magazines, TV campaigns, and social media: a chiseled man in a Paco Rabanne ad in *GQ*, a girl group in low-rise Gap jeans dancing on [TikTok](https://www.fastcompany.com/section/tiktok). “They had a person’s attention for six seconds, so they needed strong photography and a pithy [marketing](https://www.fastcompany.com/section/marketing) slogan,” says Brian Stempeck, CEO of Evertune, which helps brands increase their visibility to AI tools. AI agents, in contrast, learn about brands by reading about them on the internet. To determine the difference between similarly constructed and priced Lululemon and Vuori leggings, for example, AI will scrape e-commerce sites, product reviews, Reddit forums, and more. The problem, for brands, is that the hodgepodge of information AI cobbles together from the web might not capture all the subtleties that marketers have spent years cultivating. In an agentic future, to effectively pitch themselves to consumers, brands will need to first market themselves to AI—and do it through words. Thomas Marzano, an agentic [branding](https://www.fastcompany.com/section/branding) specialist and former head of brand at the health tech company Philips, says that brands will need to articulate in detailed language what emotions they hope to evoke, which consumers they’re targeting, their aesthetic features, their values, and more. Then they’ll need to plant this language throughout the internet, via their own channels and influencers, “so they don’t leave it to chance what the LLM will find online,” he says. Eventually, Marzano believes, brands will be able to plug this information directly into LLMs, much as they do with their inventory and pricing data. “There isn’t yet a protocol or standard to do this, but it will come,” he says. “That’s how the matchmaking between the brand and the person will happen.” But to match you with the right brand, these AI systems will also need to have a deeper understanding of your preferences. Identifying a person’s taste isn’t a new problem in Silicon Valley. Visually driven social media companies like TikTok, Instagram, and Pinterest have built complex “taste graphs,” creating a robust picture of your identity based on behavior signals like the extra second you spend on a video about Birkenstocks or whether you save a post from a luxury designer. Combined with similar data from hundreds of millions of users, these signals can predict an individual’s taste with remarkable accuracy. That’s why these social media platforms are such powerful marketing engines, connecting people to new brands—the cult sneaker, the candle startup with a three-month waiting list, the minimalist cookware. “Commerce is about picking up on subtle signals that you subconsciously collect along the way,” says Amber Atherton, a partner at the VC firm Patron who invests in early-stage startups, including ones in the agentic commerce sector. “That’s what’s missing from these AI shopping platforms.” Of all the frontier models, Gemini seems best poised to nail taste. While ChatGPT and Claude have access only to the personal information a user shares in conversations, Google has years of data it can tap if the user gives it permission. In March, it introduced a new feature called Personal Intelligence that allows users to connect Gemini to Gmail, Drive, YouTube, and Google Photos, providing it with access to receipts, vacation snapshots, and emails from brands. Over time, Srinivasan says, this will allow Gemini to move from simply responding to your shopping queries to suggesting products. This new reality will likely mean a much more personalized shopping experience. The taste-shaping technology we live with—the TikTok algorithm, the Instagram feed—was never actually designed just for you; it was created to encourage mass engagement. Over time, this has flattened taste, making everything from brand logos to coffee shops to homes look more homogenous. The next generation of AI commerce will realize that I am drawn to French minimalism and unmoved by Italian maximalism; I will spend $300 on a blouse but balk at a $50 candle; I will always pick Sézane over Reformation. For forward-thinking brands willing to make themselves legible to machines, agentic commerce could be a game changer. And for a certain writer looking for a garden-event-worthy dress, it could mean entering a world where the computer finally gets it right.