NRF 2026: AI Moves from Hype to Hard Questions

There is a specific atmosphere at retail trade conferences when a technology has cleared the hype cycle but not yet cleared the operational credibility check. Not cynical, exactly. More like a room full of people who have all read the same press releases and are quietly deciding how much of it they believe.

NRF 2026, held at the Javits Center in New York over 11–13 January, had that atmosphere.

Which is not to say nothing happened. Quite a lot happened. The show remains enormous: tens of thousands of attendees by prior-year precedent, wall-to-wall booth activations, the usual sensory overload. And the headline moment was genuinely significant. Sundar Pichai took the main stage on the opening morning to announce Google's Universal Commerce Protocol, which is either the most important open infrastructure announcement in agentic commerce to date or a very well-branded standards document, depending on your level of scepticism. Possibly both.

The Universal Commerce Protocol

The UCP is designed to give AI agents a standardised way to interact with merchant commerce systems: discover products, access inventory, process payments, complete checkout, all through a common interface that different AI systems can use without bespoke integration work per platform.

Google built it in collaboration with Shopify, Etsy, Wayfair, Target, and Walmart, with more than 20 additional retailers endorsing at launch. It is designed to be compatible with existing protocols (Agent2Agent, the Agent Payments Protocol, MCP) rather than replacing them. The "invisible shelf" framing that Google used, the idea that your products can be purchased through any AI interface that adopts the protocol without the consumer ever loading your website, is exactly the kind of paradigm shift that sounds enormous in a keynote and takes three years to actually land in production environments.

Microsoft, in a typically Microsoft response, announced Copilot Checkout at roughly the same time, which does something broadly similar within the Microsoft ecosystem. Whether UCP becomes the dominant open standard or whether the industry ends up with another fragmented protocol landscape is an open question. The early signatories looked serious, and the subsequent joining of Amazon, Meta, Microsoft, Salesforce, and Stripe to the UCP Tech Council in April 2026 suggests the industry is leaning toward convergence rather than competition. "20 retailers have endorsed it" and "it is actually in production driving meaningful transaction volume" are quite different things, of course. But the direction of travel is clearer than it was a year ago.

The Mood Shift

What was different about NRF 2026, compared with NRF 2025 where Salesforce used the show to push its Agentforce for Retail launch, was that the questions being asked in sessions and on the floor were operational rather than conceptual. Not "what is agentic commerce?" but "what does it actually do to my margin when an AI agent is optimising for lowest price?" Not "should we be building AI into the customer experience?" but "the team has been building AI into the customer experience for 18 months and the results are mixed. What are we missing?"

That is a healthier conversation. The shift from enthusiasm to implementation reckoning is not the same as disillusionment. It is the industry figuring out where the actual value is, which requires getting specific about what is working and what is not.

The AI that seemed to land best in practitioner conversations was not the flashy agentic checkout material. It was the unglamorous operational layer: computer vision for inventory accuracy, predictive demand planning that has been quietly saving retailers' margins in a period when tariff disruption has made traditional forecasting models unreliable, workforce scheduling that actually adapts to foot traffic patterns. The stuff that does not get a keynote slot.

The Physical Retail Tension

There is an interesting tension at NRF every year between the conference's roots in physical retail and the increasing dominance of digital commerce discussions. In 2026 it was more pronounced than usual, and the post-show coverage suggested this reflects something real in the industry rather than a conference-programme quirk.

The return to human-centred retail was a prominent theme, with multiple sessions arguing that the differentiation for physical retail in an AI-mediated commerce world is the thing AI cannot fully replicate: context, warmth, genuine expertise, the ability to read a customer and respond in ways that do not fit a decision tree. That is not an anti-technology argument. It is retailers trying to identify where their actual competitive advantage sits.

Smart carts, computer vision checkout, and autonomous store technology were all present on the show floor (Caper Carts, Mashgin, various Zippin implementations) but the conversations around them were more measured than in previous years. Post-show analysis suggested the economics of autonomous checkout work reliably only above a certain store size and above a certain shrinkage rate. That is a useful corrective to the tendency to evaluate technology by reference to its best-case implementations; most UK grocers, for instance, operate formats where neither condition is reliably met.

What the Show Told Us

The NRF mood gauge is worth taking seriously because the retailers who attend are not the ones still deciding whether to take AI seriously. They are the ones actively building and deploying, and the questions they are asking reflect where the bleeding edge actually is.

At NRF 2025, the bleeding edge question was: what can agentic AI do? At NRF 2026, the question was: what does it do to the customer relationship, the margin structure, and the operational workflows built over years? Those are harder questions. They are also the right ones.

The Universal Commerce Protocol, if it achieves broad adoption, changes the distribution architecture of commerce in ways that are genuinely difficult to model. The retailers who were at the show to understand it rather than just to announce it were probably getting more value from the conference than the ones with booths.

For UK retailers watching from the outside, the UCP question is not whether to adopt it eventually but when the protocol reaches sufficient adoption that ignoring it becomes a competitive risk. Given the April joiners, that timeline may be shorter than it looked in January.