Jul 7, 2026 | 8 minute read
AI is coming for shopping. But tech players are aiming at the wrong target — and B2B is about to prove it.
written by Bryan House
For the past two years, the biggest players in tech have been locked in a race to own the future of shopping. OpenAI rolled out shopping features in ChatGPT, then pulled back, then reintroduced them in a different form. Perplexity launched a buy button, and subsequently withdrew. Amazon built "Buy for Me" and watched retailers howl. Google redrew its shopping surfaces around AI answers. Everyone is fighting, everyone is pivoting, and nobody has a clear win.
And they're all focused on the wrong problem.
The conversation about agentic commerce has almost entirely centered on consumer discovery and checkout, that is, how AI can help someone find a gift, compare mattresses, or click "buy" through a chatbot instead of a browser tab. That's a reasonable place to look. It's also the hardest possible place to start, and possibly the last place AI will actually deliver value.
Meanwhile, the boring, unglamorous, enormous world of B2B procurement is sitting largely unexamined. That's where agents will actually break through first. And when they do, it will force everyone to rethink what agentic commerce is really about.
When AI rewrites a checkout flow, it doesn't inherit a blank slate. It inherits years of painstaking optimization. Brands have spent enormous resources perfecting the moment between "I want this" and "I bought this." Every upsell, cross-sell, data capture moment, and loyalty touchpoint exists because brands worked out, conversion by conversion, what moves a buyer forward and what sends them away.
An AI agent collapses all of that. The carefully tuned product detail page becomes irrelevant when a language model reads the title and price and moves on. The email capture at checkout disappears. The "customers also bought" recommendation engine gets bypassed. The post-purchase engagement sequence gets cut. In exchange, the buyer gets marginally less friction on an already functional experience.
The tradeoff isn't obvious until you look at early data. Walmart reported a meaningful drop in conversions when agents intermediate the buying experience. Amazon's "Buy for Me" triggered backlash from retailers, not because agents were bad at buying, but because agents were buying in ways that optimized for buyers at the explicit expense of sellers.
The consumer AI commerce problem is also a trust problem in the wrong direction. Commentators keep arguing that consumers won't trust AI to make purchases on their behalf. I don't buy that argument. Consumers trust Temu with their credit card and their data despite well-documented concerns about where that data goes. Trust has never been the real gating factor in consumer adoption — convenience and value are. The blocker is that nobody has figured out a consumer AI commerce experience that clearly beats the existing one.
B2B buying is not a smoother version of consumer shopping. It's a completely different process, organized around different constraints, and almost none of those constraints create a good experience. They exist to manage complexity.
Consider what a mid-market manufacturer navigates to purchase something as routine as industrial fasteners. There are approved vendor lists, negotiated contract pricing that differs by account, compatibility requirements across product lines, multi-location delivery splits, approval workflows that vary by order size, PO generation that touches the ERP, invoice reconciliation, and, somewhere buried in all of it, a question about whether the specific screw they need is actually in stock at the distribution center closest to the factory floor that needs it.
That process currently runs on a combination of sales reps, spreadsheets, email chains, and institutional memory. It is slow, expensive, error-prone, and completely dependent on humans who are expensive to hire and difficult to retain. According to Sequoia Capital's recent analysis of where AI will create the most value, the highest-opportunity targets are processes that are intelligence-heavy but not judgment-heavy — meaning they involve navigating complex rules, not making genuinely ambiguous strategic decisions.
B2B procurement is almost entirely intelligence-heavy. Things like compatibility, pricing tiers, approval thresholds, entitlements, minimum order quantities, lead times, and regional availability are all rules. The sales rep navigating this process is not exercising some sort of deep strategic judgment. They are applying institutional knowledge to a decision tree that, if anyone had taken the time to encode it properly, would run itself.
That is exactly where AI agents win.
There is a counterintuitive case to be made that B2B will lead a major technology transition before consumer technology does — and that has almost never happened in the history of the internet.
The pattern has always run the other direction. Consumer products achieved massive scale, then enterprise versions followed. Search was Google before it was enterprise knowledge management. Social networking was Facebook before it was Slack. Mobile was the iPhone before it was mobile device management. Consumer led; enterprise followed.
Agentic commerce may flip that pattern for a specific reason: B2B organizations have better data.
Not better in the sense of cleaner or more thoughtfully organized — most B2B product catalogs are a disaster of PDFs, legacy ERP exports, and specs that live in salespeople's heads. But better in the sense of more structured, more rule-bound, and more amenable to the kind of explicit encoding that AI agents actually need to function reliably. A consumer brand's product data is a set of images, a description, and a star rating. A B2B distributor's product catalog is a relational system of SKUs, compatibility matrices, account-specific pricing layers, and contractual constraints. The consumer data looks richer. The B2B data is actually more useful to a machine.
Add to that the scale of the opportunity. B2B e-commerce globally is roughly five to six times the size of B2C e-commerce by transaction volume. The friction in B2B procurement is a core cost of doing business, and it is enormous. A company that eliminates even a fraction of that friction doesn't slightly improve a checkout experience; it restructures a category.
None of this works without confronting the real infrastructure problem underneath both B2B and consumer AI commerce: product data is not built for machines.
Commerce infrastructure has been optimized for presentation. We built better product information management systems. We built richer catalogs. We built product pages that look excellent on a retina display. And we built all of it for humans to interpret.
An AI agent doesn't interpret. It evaluates. It needs to know what is compatible, what is allowed, what is in stock, what the price is for this specific buyer under this specific contract at this specific moment. If that information isn't explicit, structured, and accessible, the agent doesn't ask a clarifying question. It moves on.
This is almost exactly the problem enterprise data faced a decade ago. Companies were drowning in data they couldn't use because the data was organized around how systems stored information, not around how decisions actually got made. The breakthrough came from building semantic layers — encoding meaning explicitly, making relationships machine-readable, creating stable abstractions that both analysts and systems could reason against.
Commerce needs to make the same move. And the companies that do it in B2B will have a significant head start, because B2B product relationships are already more explicit. Compatibility is not a matter of taste. Pricing is not vibes. The rules are there — they just need to be encoded.
The protocol wars consuming the tech industry's attention — who controls the agent handoff layer, whether OpenAI's shopping approach or Anthropic's or Google's becomes the standard — are a real fight over a real prize. But they are also a distraction from the more foundational question: who will be legible to agents when those standards settle?
Legibility is the new distribution. If an agent can't confidently evaluate your product, reason about your pricing, and execute against your entitlements and constraints, you are invisible because the system can't work with you.
B2B merchants who structure their product data for machines, encode their business rules explicitly, and build integrations that agents can call reliably will not just survive the transition. They will be the ones agents default to, because they are the ones agents can actually use.
Consumer commerce will get there too. But it will take longer, cost more, and require breaking a lot of things that currently work before building something better. B2B procurement doesn't have that problem. Most of it is already broken. AI doesn't have to replace something good. It just has to be better than a fax machine and a sales rep eating lunch.
That bar is much lower than anyone seems to realize. And clearing it will be worth a lot more than fixing checkout.
Schedule a demo to see how Elastic Path delivers unified commerce for leading global brands.