The protocol race
Retail tech had two big projects running in parallel this year. One built checkout infrastructure for a shopper that barely exists yet: the autonomous AI agent. The other kept trying, quarter after quarter, to make money on a shopper who already exists and orders constantly: the person tapping "deliver in ten minutes" on a phone. The second project is the harder one, and it shows.
OpenAI and Stripe released the Agentic Commerce Protocol on September 29, 2025, an open standard, licensed under Apache 2.0, for letting an AI agent hand a merchant a scoped payment token and complete a purchase without the shopper leaving the chat interface. It launched alongside ChatGPT's Instant Checkout feature, live with Etsy on day one and later Shopify brands including Glossier, Vuori, Spanx and SKIMS. PayPal joined as a payment provider on October 28, 2025.
Google answered in January. Sundar Pichai announced the Universal Commerce Protocol at the National Retail Federation's NRF 2026 keynote on January 11, 2026, co-developed with Shopify and backed by more than 20 retailers and payment processors, including Etsy, Wayfair, Target and Walmart. UCP is now wired into AI Mode in Google Search and the Gemini apps, letting shoppers buy from participating retailers without leaving the chat. Bain and Company has projected that agentic commerce in the United States alone could reach $300 billion to $500 billion by 2030, a number retailers have been repeating in almost every panel discussion since.
The agent that didn't show up
Then, on March 4, 2026, OpenAI quietly pulled Instant Checkout. In its own account, the company said the feature "did not offer the level of flexibility that we aspire to provide," and it is now steering merchants back toward their own checkout flows while ChatGPT focuses on product discovery instead. Coverage from CNBC and Forbes at the time noted the underlying problem was more basic than flexibility: OpenAI struggled to onboard merchants, could not reliably show accurate product data, and never built support for multi-item carts or loyalty programs. Fewer than 30 of Shopify's millions of merchants had gone live with it in the six months it existed.
That is a rough result for the feature that the industry pointed to as proof agentic commerce had arrived. It does not mean the protocols are pointless. ACP and UCP still matter as plumbing, and Google's version has real retailer commitments behind it. But it is a useful correction to a year of retail conferences that treated "the AI agent will do your shopping" as close to inevitable. The infrastructure got built faster than the demand for it did.
The shopper who is already there
Meanwhile, the customer retail tech already has, the one ordering groceries and snacks for ten-minute delivery, is still expensive to serve in most cases. India's quick commerce sector illustrates the gap plainly, because its three largest players report results on a similar timeline. Blinkit's adjusted EBITDA turned positive at 102 crore rupees for the quarter ending June 2026, its fifth straight quarter of margin improvement, while adding 200 net new dark stores to reach 2,443 nationwide, according to parent company Eternal's results. Zepto has not made that turn. Its updated draft prospectus, filed with India's securities regulator ahead of a planned IPO, shows a loss of 5,905.19 crore rupees for the 2026 fiscal year, up 26 percent from the year before, even as revenue more than doubled to 22,624 crore rupees. Swiggy's Instamart posted a consolidated net loss of 1,092 crore rupees in the July to September 2025 quarter, wider than the 626 crore rupee loss a year earlier.
Three companies, the same delivery model, three very different cost structures. One of them figured out density and store economics well enough to stop losing money on each order. The other two are still buying growth with cash, and the DRHP filing shows regulators and investors are now the ones checking the math.
What this says about the year
Retail tech spent 2026 building standards and press releases for a shopping agent that mostly failed its first real test. It spent less visible effort on the harder, less glamorous problem of making the existing delivery model pay for itself, and the difference between Blinkit and Zepto shows that problem is solvable, just not automatically. The lesson for anyone watching this category is not that agentic commerce is a dead end. It is that infrastructure for a hypothetical customer is cheaper to announce than margin discipline is to execute, and retailers should be more skeptical of the first than they are of the second.