Why the AI shopping agent wars will heat up in 2026

That dilemma is what led Bornstein to create Daydream, an AI-powered shopping platform focused on fashion. Daydream, which is still in beta, launched in June, a year after the company announced $50 million in seed funding. The platform lets users describe what they’re looking for in conversational language — or by uploading a photo — and receive personalized product recommendations in response.

Daydream is one of a growing number of tech startups, retailers and brands racing to hand off parts of the shopping journey to AI-powered agents — tools designed to research, recommend and, in some cases, buy products on a consumer’s behalf. In 2025, nearly every major retailer or AI incumbent entered the race for “agentic” commerce. Amazon expanded its Rufus shopping assistant with a new automatic-buying feature; OpenAI embedded checkout directly into ChatGPT; and Perplexity rolled out an AI-powered browser that prompted legal pushback from Amazon over how it accessed its site.

If 2025 was the year that companies laid the groundwork for agentic commerce, then 2026 will be the year retailers, tech giants and startups jockey to determine whose AI agent becomes the default interface for shopping — if consumers even adopt these AI agents en masse at all.

“We’re pre–Sputnik launch phase,” said Juan Pellerano-Rendón, chief marketing officer at e-commerce software startup Swap, which launched a new platform in September that creates AI shopping agents for brands to offer on their sites. “Everyone is building the spaceship, but no one has really launched it yet.”

Winning the buy box

Perplexity got a head start in the agentic shopping race when it unveiled an AI shopping agent for its U.S. subscribers in November 2024 that lets paying users research and buy products directly within Perplexity’s app or website. 

But throughout 2025, new AI shopping agents sprouted left and right. Since September, OpenAI has announced several deals with major retailers — the latest partners include Target, Instacart and DoorDash — to let shoppers buy from their platforms directly within ChatGPT. Amazon released an agentic tool called “Buy For Me” that lets consumers shop other brands’ and retailers’ websites without leaving Amazon’s app. Even brands, including luxury players like Ralph Lauren, are launching their own AI-powered shopping assistants.

Alongside the big tech players, a wave of niche AI shopping startups emerged in 2025. New apps like Phia, a price-comparison platform co-founded by Phoebe Gates, and OneOff, which recommends products based on celebrity and creator looks, are targeting specific shopping use cases rather than trying to do everything at once. Even as these startups face competition from AI giants like OpenAI, investor interest remains strong. Phia, for instance, raised $8 million in September.

It’s still early days, and for the time being, shopping only accounts for a fraction of overall AI use. According to a working paper authored by OpenAI’s Economic Research team and a Harvard economist, around 2% of all ChatGPT queries are shopping-related — things like “recommend a laptop under $1,000” and “how much are running shoes.” Still, that comes to around 50 million shopping queries daily. Even a sliver of that activity could translate into enormous volume at scale.

The allure of agentic shopping is easy to understand. Online shopping has become overwhelming, with millions of products spread across multiple retailers, marketplaces and DTC sites, plus endless filters, ads and reviews to sort through. An AI agent, in theory, promises a shortcut: Instead of clicking around, shoppers can describe what they’re looking for in a sentence or two and get a short list of options that fit their needs. The idea is that the agent handles the busywork — weighing price, availability and quality — so shoppers can move more quickly toward a purchase.

To win the AI agent shopping race, platforms will need to win over customers. To some extent, that may give companies like OpenAI and Amazon an advantage because they can introduce agentic commerce features inside products consumers already use habitually. OpenAI announced in October that ChatGPT has reached 800 million weekly active users, up from 500 million in March. Meanwhile, Amazon captures 40% of all U.S. e-commerce spending, and 90% of its shoppers begin their searches directly on Amazon.com.

Scot Wingo, author of the Substack Retailgentic and founder of ReFiBuy, a company that helps brands and retailers optimize for agentic AI, said the quality of agents’ product recommendations will ultimately determine which platforms gain consumer trust. “GenAI knows far more about the shopper than Google ever did; the job now is to expand and contextualize the product catalog so the engine can map that shopper context to the right SKU,” he said. “There’s a big gap between tight, keywordy product catalogs and the context GenAI needs.”

In other words, shopping agents are only as good as the data they can access — and that’s where things get complicated. To make accurate recommendations, agents need more granular and standardized information — not just basic attributes, but detailed descriptions of fit, quality, durability, shipping timelines and inventory status. For decades, merchants have structured their product catalogs around search engine optimization, designed to match short, keyword-based queries rather than long, descriptive requests.

Shoppers now give AI agents far more context — dozens of words about who they are, how they’ll use a product and what they prefer — but most product catalogs were never built to absorb or respond to that level of detail. Closing that gap would require brands to add far richer context to every product, a labor-intensive process that goes well beyond basic fields like price or size. And some of the information agents would need is considered proprietary, so some merchants may be reluctant to share it with third-party AI platforms.

“Even if these agents don’t bother with the transaction and just offer themselves up as an alternative to Google search, there is still a lot of creating new information in new formats for each product that every merchant will need to do,” said Sucharita Kodali, a principal analyst at Forrester. “There will no doubt be holdouts.”

That tension is already shaping how the agentic shopping market is developing. Unlike many other retailers that have inked partnerships with third-party AI companies like OpenAI, Amazon has blocked off its site from outside agents. The move was likely intended to protect its $56 billion advertising business, which depends on shoppers browsing Amazon’s site, Modern Retail previously reported. Earlier this year, Amazon sued Perplexity, arguing that the AI startup improperly accessed its site, a legal fight that underscored how high the stakes have become as platforms jockey to control the future of AI-driven commerce.

But while the rise of AI shopping agents from ChatGPT to Daydream challenges Amazon, the e-commerce giant doesn’t need to worry about its retail empire being upended just yet, according to Kodali. That’s because Amazon already has stored payment credentials, consumer trust and habitual usage. In a July survey of 700 consumers conducted by Forrester, only about a third said they would be willing to complete payment through an answer engine at all, citing data privacy concerns.

To gain traction with consumers, and compete with the likes of Amazon, AI companies like OpenAI “need to get shoppers on board and convince them that it’s better to buy on ChatGPT than on a retailer’s website, which usually a third party can only do if the assortment is bigger or if the products are cheaper,” Kodali said. “So long as Amazon is a player in e-commerce, I don’t see ChatGPT ever winning the buy box over Amazon.”

Not the finish line

Building an AI agent specifically for shopping is no easy feat. Users shop with nuanced, conversational requests — “a dress for a wedding in Paris in the fall,” for example — which involve interpreting multiple factors — occasion, seasonality, location, style, color, fabric, price point and more, sometimes in a single query, Bornstein said. Standard LLMs struggle to reliably extract all pertinent details at once, often omitting or misinterpreting elements.

Indeed, Amazon Chief Executive Officer Andy Jassy recently told analysts that most AI shopping agents fail to provide a satisfactory customer experience. He said most lack personalization and often provide inaccurate pricing and delivery estimates. “We’ve got to find a way to make the customer experience better and have the right exchange of value,” he said in October. Bloomberg reported that product recommendations from Amazon’s Rufus, OpenAI’s ChatGPT and Walmart’s Sparky elicited fairly generic results when asked what to buy mom for Christmas.

Rather than relying on keyword searches or static filters, Daydream uses multiple AI models to interpret factors like occasion, seasonality and personal style. For those reasons, AI shopping startups with a specialized focus like Daydream may have a competitive edge against more mainstream AI tools heading into 2026. Bornstein declined to share revenue figures, but she said the platform’s catalog includes nearly 2 million products spanning more than 10,000 brands.

Ultimately, the biggest hurdle facing any AI agent will be consumer adoption, according to Swap’s Pellerano-Rendón. “There’s no innovation without utility,” he said, arguing that agents won’t improve unless consumers see enough value to keep using them. Right now, he said, the industry is stuck in a “chicken or egg phase,” where “the agent needs to get better, but also the consumers need to use it for it to get better.” Until shoppers consistently experience moments where an agent truly understands what they want — what he described as a “needle in the haystack feeling” — adoption will remain slow. “We’re still super, super early days, in terms of how people are using agents to shop,” he said.

Bornstein agreed.

“It’s going to take longer to perfect these experiences than anyone thinks,” she said. “It will happen, but 2026 will be a step along the way, not the finish line.”