The last four briefings all pushed the same point: be readable or be skipped. This cycle the big commerce platforms started shipping readability themselves, as a built-in feature, which quietly ends that as an advantage. The gap that replaces it is narrower and much emptier.
On July 27, Adobe put agent-readable product data into Adobe Commerce as a standard feature. Not an add-on, not a partner integration. A Catalog Agent that rewrites your product names, specs, availability, pricing and purchase rules into the form a machine wants, and publishes it over Google's Universal Commerce Protocol so any AI shopping agent can consume it.
Shopify, Salesforce, Dynamics and commercetools already had their own version of that path. Adobe joining them means the single most repeated piece of advice in this market, make your catalog readable to agents, is on its way to being something your software vendor just does for you. Adobe's stated reason was blunt: AI traffic to US retail sites ran 125 percent above last year between April and June.
When a thing becomes free, it stops being a difference. It becomes the floor.
The advice everyone was giving is turning into a checkbox.
Think about what that does to the shape of the market. A year ago, publishing clean machine-readable product data was work: a project, a budget line, a small advantage over the store down the road that hadn't done it. Now, if you sit on one of the large commerce platforms, it arrives in a release note. You will be readable because your vendor decided you should be.
That is good news and it is worth taking. It also resets the question. Once every store on a major platform can be read cleanly, an agent comparing options is not choosing between readable and unreadable anymore. It is choosing between the stores it can finish a job at and the ones where it hits a wall halfway through.
Last cycle the shelves got graded. This cycle the graders started shipping the fix, and the frontier dropped a layer.
Our last briefing covered an audit that scored 141 product pages at 29 leading retailers and found seven in ten failing on machine-readable basics. That finding still stands. What changed in a week is who is expected to fix it. Adobe answered for its own customers, and the other platforms had already answered for theirs.
Meanwhile an independent crawler published a count that reframes the whole problem. Of 11,414 stores it verified as publishing the UCP standard, nearly all publish a checkout link, and about 10,100 publish cart and order endpoints too. Fifteen publish a way for an agent to sign a customer in. None publish a way for an agent to pay them directly.
Across 11,414 stores verified as publishing Google's Universal Commerce Protocol. Browsing an agent can do almost anywhere. Signing your customer in, almost nowhere.
Source: UCP Checker, state of agentic commerce, July 2026. An independent crawler rather than a research house, so read the shape of it rather than the exact counts. The shape is not close.
Put plainly: agents can window-shop almost everywhere and finish almost nowhere. A payments piece that ran widely on July 31 made the same argument from the other end, that the real contest in agentic commerce is not how autonomous the agents get, it is authorization. Who counts as a verified agent, what proves the shopper actually asked for this, and who eats it when a purchase goes wrong. Those answers decide which businesses agents transact with and which ones get pushed into a manual queue that no agent waits around for.
The volume arrived before the plumbing did.
None of this would matter if agent-driven buying were still a demo. It isn't. Alipay's AI Pay crossed 120 million transactions in a single week, the first agentic payment service to hit that scale. Salesforce reported agentic search, shoppers who start by asking an assistant instead of typing into a search bar, up 200 percent year over year. In the same research, retailers running their own shopping agents grew sales 59 percent faster than the ones sitting it out, while only 28 percent of commerce organizations use agentic AI today.
The payment industry spent late July building for this in public. MoonPay launched a vault that lets ChatGPT and Claude users authorize card transactions without the agent ever seeing the card. Corpay started issuing virtual cards to agents. Zero Hash launched a suite for agent-driven money movement. All three inside four days. Visa and Mastercard both named agentic commerce a growth driver on their quarterly calls, and PayPal stood up a division with it listed as a growth area.
If you sell to other businesses, the number to sit with is Gartner's: about 90 percent of B2B purchasing, roughly 15 trillion dollars of it, running through AI agents by 2028. B2B ordering is where custom-built systems are the norm and where no platform vendor is racing to hand anyone a Catalog Agent. Travel is moving the same way, with Amadeus signing on as a founding UCP partner for lodging and Radisson launching hotel search and booking inside ChatGPT.
Regulators noticed too. New York's financial regulator is working through how consumer protections apply when a purchase self-executes with no human approval, naming chargebacks, disputes and liability as the exposure, and Delaware proposed a legal framework for autonomous agents the same week. Whichever way those land, they push in one direction: being able to show who authorized what, with limits you set, stops being a nice-to-have.
A separate, related free tool
One thing worth checking before any of this: whether agents can reach your site at all. Our free AI Visibility Checker writes you a correct robots.txt so the crawlers you want get through and the ones you don't stay out. It matters most if you're on Cloudflare, whose defaults change on September 15, 2026.
Set your AI crawler rules →No sign-up, about two minutes, and it doesn't touch your search ranking.
The sign-in and the limits are the product, not a phase two.
The reason so few stores publish a customer sign-in is not that nobody thought of it. It is that publishing one means letting a piece of software act inside your accounts, and no sane business does that without deciding first exactly what it may touch. That decision is real work, and it is the work we do.
BeaconSpec takes the REST and GraphQL APIs your business already runs and turns them into one curated server that agents discover and use over UCP, published at your own domain. You pick exactly which operations are exposed, so nothing goes out that you haven't approved. The standard login handshake is handled for you, so an agent signs the shopper into your existing accounts, with no shared passwords and no separate build for each AI. If your rules say the whole thing has to run inside your own walls, you can host it yourself.
Read the two halves of that chart again. The green half is what your platform vendor will hand you, or already has. The red half is what somebody has to build against your actual systems, because a sign-in and a spending limit cannot be inferred from a product page. Your order system already knows who your customer is and what they're allowed to buy. It just has no way to say so to a machine.
Being findable is becoming a feature you're given. Being transacted with, on your terms, stays something you decide to build.
Readable is turning into table stakes. Buyable is still wide open.
Of 11,414 stores on the agent standard, nearly all can be browsed and 15 can sign a customer in. If your competitors are about to be handed the first half for free, the second half is the only half left that decides who gets the order.
This is a market briefing we run every cycle, tracking what's actually changing in agentic commerce so you don't have to piece it together yourself.
Four briefings ago the argument was that agents were starting to choose. Then the payment rails went live. Then someone graded the shelves and most stores failed. Now the platforms have started fixing the grade for their own customers, and the thing agents cannot do at almost any store is the thing that finishes a sale. If you build on your own stack, nobody is coming to close that for you, which is the bad news and also the reason the opening is still there.
BeaconSpec exists for exactly this shift: making your business discoverable and transactable by AI agents over the UCP standard, instead of invisible to them.
See what we do • Read more articles
Related reading: Someone finally graded the shelves. Seven in ten stores failed.
David Soden writes about agentic commerce, automation, and building durable technical systems for businesses. Photography via Pexels (Susanne Plank, Caner Kökçü, Kampus Production); figures cited are drawn from the public reporting named above.