Shop ’til you bot: Google, OpenAI, and the race to build agentic commerce

Last September, OpenAI and Shopify made an announcement that sent ripples throughout the retail industry: They were partnering to launch Instant Checkout—a feature that would let people complete purchases directly within ChatGPT.

Within months, the AI giant promised, we would be able to ask ChatGPT for Mother’s Day gift ideas or top-rated lightbulbs, and then click to buy products instantly. Shopify’s president, Harley Finkelstein, declared this the “the new frontier” of retail.

But if you’ve tried to shop on ChatGPT recently, you know that this future never arrived. OpenAI quietly killed Instant Checkout in March. The official story, according to OpenAI’s blog post, was that the checkout feature “did not offer the level of flexibility that we aspire to provide.” 

The unofficial story is that OpenAI and Shopify were unprepared for the level of complexity checking out requires. Fewer than 30 of Shopify’s millions of merchants ever went live.

This is the state of AI shopping in 2026. The same company reportedly being trained to guide drone strikes in active conflict zones cannot build a working check-out.

My interviews with executives at Google, OpenAI, Stripe, Walmart, and a long list of AI-focused startups, revealed that the technology powering AI—large language models—is incompatible with existing e-commerce technology.

Behind the scenes, there’s a massive effort underway in the retail industry to build the infrastructure required to make AI shopping possible. The commerce leads at Google and OpenAI, the two biggest players in the space, say that we’re months—not years—away from a tipping point where agentic commerce really will become commonplace.

Whoever makes the shopping experience consumers want to use will own one of the most lucrative pieces of real estate in the history of retail.

In this story you’ll learn:

• The knotty problem that forced OpenAI to pull back on instant checkout
• How frontier labs are rebuilding commerce infrastructure from the ground up
• Which company is most likely to win the AI shopping war

OpenAI’s false start and new vision

Last year, as AI became the fastest-adopted technology in history, AI companies realized they needed to turn their attention to commerce. There’s a lot of money hanging in the balance. According to McKinsey, AI-driven commerce could generate $1 trillion in U.S. revenue and up to $5 trillion globally by 2030.

As we entered 2026, the retail industry’s newest buzzword was “agentic commerce,” which refers to AI agents shopping autonomously on the user’s behalf. “Nobody has figured it out, but everyone has FOMO,” says Emily Pfeiffer, principal analyst at Forrester, who covers AI and commerce. “Everyone is prematurely rushing to market.”

Case in point: The botched Instant Checkout roll-out. When they announced the checkout feature, Shopify and OpenAI had promised that millions of Shopify merchants would soon be shoppable from ChatGPT, alongside Etsy sellers and Walmart. A tiny fraction of the integrations were built.

“Shopify has major egg on their face,” says Omar Qari, the CEO of Logicbroker, which helps brands feed product data into LLMs. “If you go back to late last year, OpenAI said, ‘We’ve solved it. We’re connecting all the world’s products inside ChatGPT and it’s going to be an amazing shopping experience.’ But they literally couldn’t even get it live.” (Shopify declined to comment on this story.)

Neel Ajjarapu, who leads commerce at OpenAI, admits that building a checkout was more complex than the company had envisaged, and ultimately, merchants were best positioned to build these tools.

“It’s not enough to have a basic checkout page,” says Ajjarapu. “You need to think about things like loyalty points, in-store pickup, basket promos, and dozens of features that are specific to the geography, category, and merchant type.”

Rather than try to build all of that from scratch, OpenAI decided merchants should own checkout themselves. “Merchants are already optimized this part of the funnel,” says Ajjarapu. “We’re going to make it really easy for them to bring that into ChatGPT.”

But even without Instant Checkout, consumers are already turning to ChatGPT for their shopping needs: According to Pew, roughly 2% of queries to the chatbot—about 50 million a day—are shopping related. Ajjarapu says that ChatGPT is good at helping users figure out how to buy products that require a lot of research, like electronics, appliances, and sports gear. OpenAI has every intention of transforming ChatGPT into the world’s personal shopper.

“The goal is for ChatGPT to be a super assistant,” he says. “When it comes to shopping, it should be able to help you find products, optimize carts, and buy things. It will be able to help you discover things that you never knew before, but are tailored to your personal circumstances.”

The Google advantage

Predicting taste is precisely where ChatGPT is at a disadvantage. The chatbot only has access to information you share in conversations to tailor product recommendations to your needs and taste. But its biggest competitor—Gemini — has access to a much deeper trove of knowledge about you thanks to all the information you have shared with Google over the years.

In March, Google rolled out a feature called Personal Intelligence, which lets Gemini’s 750 million active users tap into their data in Gmail, Photos, and Drive when answering queries. Once you give Google the permission to access this data, the model will know everything from your travel plans to which brands’ marketing emails you open. According to Google’s early projections, 75 million users had activated the feature.

“Once the model has the opportunity to learn about you, it can start at a better point than starting from scratch and expecting that you will tell us everything about you,” Srinivasan says. “Because it is much easier to give Gemini five pictures of clothes you like than to describe your dressing style.”

Without years of data about the people using ChatGPT, OpenAI is in a much weaker position. It must gather crumbs from the details you happen to share in your conversations.

“ChatGPT is starting to learn so much more about you as a user, not just in your retail taste, but everything else happening in your life,” says Ajjarapu. “We can start using that data to help make extremely well-personalized recommendations that match your taste.”

But user data isn’t Google’s only advantage. It also has better access to product data. The Shopping Graph—Google’s real-time database of product pricing, inventory, and merchant relationships—traces its origins back to Froogle, a shopping platform that Google launched in 2002. That graph has been refined, expanded, and integrated with every part of Google’s ad and merchant infrastructure ever since.

“We’ve had decades of experience with the Shopping Graph,” says Vidhya Srinivasan, Google’s vice president and general manager of advertising and commerce. “We’ve invested a lot in having the repository of products that has the diversity of merchants, but more importantly, it’s updated every second, every minute of every day.”

Building the plumbing of agentic shopping

One reason AI companies have struggled so much to build shopping tools is that their underlying technology wasn’t designed for commerce. Large language models are trained by scraping the entire textual archive of the internet—learning to predict the next word in a sentence or the next fragment of code. This has proven remarkably effective for drafting a college term paper or writing an app.

But a product page is different from a web page. A lot of crucial product information—like inventory, shipping costs, and when it launched—does not appear on websites. “OpenAI’s first attempt at trying to get products into ChatGPT was to screen-scrape Dick’s Sporting Goods or Ulta and show their products,” says Qari. “And you can’t blame them, because that’s how they trained the model.”

AI companies now realize they need to build the plumbing for agentic commerce from scratch. There’s currently a race to create a new standard to make retailers’ real-time product data readable by LLMs.

OpenAI and Stripe co-developed the Agentic Commerce Protocol (ACP), which they open-sourced last year. Google, Shopify, and a coalition of two dozen retailers and payment companies—including Etsy, Wayfair, Target, Walmart, Visa, Mastercard, and Stripe itself—launched the Universal Commerce Protocol (UCP) in January. UCP is more robust, since it can handle complicated things like scheduling and returns. But, for the moment, every brand and retailer is being told it needs to support both.

In a sign that Gemini is pulling ahead of ChatGPT in retail, Google has been announcing a string of new features over the last few months. Real-time pricing and inventory are already live on Gemini, as are in-chat checkout via Google Pay. As of March, Gap became the first major fashion retailer to allow shoppers to complete a purchase entirely inside the chat and in April, Ulta Beauty announced it would be doing the same thing.

What happens next

Today, shopping via chatbot means you’re ultimately shopping via the tradition web, with tabs that mushroom on your screen. But a few years from now, it’s likely that the ingredients I need for dinner will be ordered the moment I share the recipe with my AI agent.

Christmas gifts—which currently eat an entire December weekend—will be handled in the time it takes to drink a coffee: the agent knows that my mother likes gardening books, that my daughter’s best friend is obsessed with slime, that I always overspend on my husband and should probably have a hard limit.

When a wedding comes up, the agent will see it on my calendar and suggest appropriate dresses I like. Some purchases I’ll sanction with a tap. Others will simply show up, correctly, without my having asked.

The infrastructure to make this happen is being built slowly, in fits and starts, with the occasional embarrassing pullback. But there’s a lot of money on the table, which is incentivizing AI companies to pour resources into building shopping tools. Whoever builds the best agentic commerce platform is going to have the first mover advantage and lock in a generation of consumers.

Right now, the smart money is on Google. It has both the merchant relationships and deep knowledge of users, if they opt in to Personal Intelligence. And the protocol that Gemini’s checkout runs on— UCP—looks like the stronger foundation.

But as with everything in AI, things are moving fast. And OpenAI is not out of the race. And the field is changing so fast, nobody can call the winner yet.

“Every couple months, we just see such massive changes to what our models are able to do,” says OpenAI’s Ajjarapu. It is impossible for me to predict what’s going to happen on what timeline.”

View the full article