On November 13 OpenAI gave ChatGPT's agents a way to pay, and on the same day a standard arrived for how agents find a business. An audit of 200 big US retailers and brands found only 16% publish anything an agent can use. The paying part is solved. The finding part is still mostly up to you.
OpenAI launched ChatGPT Wallet on November 13 with Visa, Mastercard and American Express, CNBC reported. A shopper stores a card once, asks an agent to find something, and approves each purchase. The agent never sees the card number. The card network issues a token that only works for that one transaction. At launch the wallet is limited to US subscribers on the Pro plan and isn't available in the EU, but ChatGPT has more than 1.2 billion weekly users, and every one of them is a potential shopper with an agent that can pay.
The same day, Sierra released version 1.0 of its Personal Agent Protocol, nicknamed Poppy. It's a small file a business puts on its own domain, poppy.json, that tells personal agents which channels and capabilities they may use and how a shopper signs in. Shopify now publishes one for every store it hosts, in more than 175 countries, and Meta's Muse agent checks for it first on every site it visits, Unite.AI reported. Sierra lists more than 100 design partners, including Cisco, Delta Air Lines, Intuit, Salesforce, Twilio and Zoom.
Agents can pay now, and they know where to look for you. What they find there is up to you.
An agent shopping for someone needs two things: a way to pay, and a way to read the store.
For most of this year the payment half was the hard part. Banks, card networks and wallets spent months working out how an agent proves it has the shopper's permission. ChatGPT Wallet settles that for a large share of shoppers, using the same card networks merchants already accept. The other half is the store itself. An agent doesn't browse the way a person does, so it looks for a published, machine-readable description of what a business sells and what it can do. UCP, the open commerce standard from Google and its partners, is one such description. Poppy is now another, and Shopify merchants got both without lifting a finger.
Merchants who run their own systems got neither. Human Forward, a small research firm, checked 200 major US retailers and brands this month and found that only 16% published a UCP manifest or a storefront MCP endpoint, BikeBiz reported on November 13. In UK cycling the same check passed 70 of the 147 sites it could assess, and 16 sites blocked the checker outright. The audit confirms that the files exist, not that a purchase goes through, so treat the number as a direction rather than a measurement. The direction is clear enough: roughly five in six big US merchants publish nothing an agent can use.
A UCP manifest or a storefront MCP endpoint, per Human Forward's readiness check run November 6, 2026.
Source: Human Forward readiness report, as reported by BikeBiz, November 13, 2026. The UK figure is 70 of 147 sites, rounded. The check confirms the files exist, not that an agent purchase works, so read both bars as directional.
Last week agents were guessing. This week they got a wallet, and the guessing got more expensive.
Our last briefing covered Visa's benchmark, where AI shopping agents failed 42.8% of test runs and often bought something anyway. Those agents were working from whatever they could piece together about a store. A wallet doesn't make them read better. It makes it easier for them to finish the purchase, so the facts an agent finds about you matter more this week than they did last week.
Two gatekeepers moved as well. JPMorgan now stops AI agents from shopping with a Chase card number, on top of an October ban on agents logging in to Chase accounts, PYMNTS reported from the Wall Street Journal. Bank of America, Capital One and Citi still allow it. Stripe's Link works around the block with one-time virtual cards, though merchants say those cost more to accept and make returns harder. And Reuters reported on an internal Amazon draft for a "World Product Graph": data on nearly every product sold anywhere, fed by brands and by crawling other retailers' sites, so shoppers never need to leave Amazon. Coca-Cola and Bayer reportedly agreed to contribute and P&G is in talks. The draft targets October 2027, and Amazon calls it brainstorming.
If you don't publish what an agent needs, someone else's description of you fills the gap.
When an agent comes looking for your business and finds no published file, it falls back on something else. It can read your web pages and guess, which is how Visa's agents went wrong. It can get stopped by your bot protection and move on to a store that let it in. Or it can rely on whatever a marketplace says about your products, and Amazon's graph would make that marketplace's answer the default for a lot of agents. None of those paths leaves you in charge of your price, your stock, your return terms or what the agent is allowed to do.
Vercel told Retail Week this week that not enough retailers are ready for agentic commerce, "but it's not a crisis yet." That fits the numbers. The wallet is Pro-only and US-only, Poppy is days old, and agent order volumes are still small. Black Friday on November 27 is the first real test, and the weeks before it are a cheaper time to publish your own description than the quarter after a competitor's agent sales start showing up.
We publish your business to agents from your own systems, and keep it current.
BeaconSpec is the discovery layer for agentic commerce. You already run APIs behind your catalog, booking engine or ordering system: REST, OpenAPI or GraphQL. We turn them into one curated MCP server, the open protocol AI assistants use to call outside tools, and publish its UCP profile at your own /.well-known/ucp. That's the kind of file Human Forward was checking for, served from your domain and describing your business in your terms.
You choose exactly which operations an agent may call, and anything left off the list doesn't exist as far as the agent is concerned. When a task needs the shopper signed in, like checking an order, we handle the standard OAuth login so the agent works inside that shopper's own account. We host it, or hand you a sealed container to run in your own environment. BeaconSpec isn't a marketplace or a payment network, so no platform sits between you and the agent, and nobody else can switch your listing off.
The standards will keep moving. Poppy 1.0 is still marked draft, and Sierra says it may change in ways that break earlier versions. Sierra also ships a free kit that generates a file like this, so for a team with engineers the first version is the easy part. The work that lasts is deciding what the file exposes, running it somewhere you control, and keeping it correct each time a spec changes. That upkeep is the job we do.
AI agents can pay now, and they look for a published file to find out what a business offers. Only about one in six big US merchants publishes one.
Shopify stores got that file automatically. A business on its own systems has to publish its own, or let an agent guess, or let a marketplace describe it.
Volumes are still small, which makes the next few weeks the cheap time to do it.
Worth watching over the next month: agent purchases through ChatGPT Wallet during Black Friday week; whether the wallet opens beyond Pro subscribers; which merchants off Shopify publish a Poppy file, and how the spec changes after 1.0; whether other banks follow JPMorgan; and which brands sign up to feed Amazon's product graph.
BeaconSpec exists for exactly this shift: making your business discoverable and transactable by AI agents over UCP, on an interface you publish and control, kept correct as the standards underneath it keep changing.
David Soden writes about automation, web strategy, and building durable technical systems for businesses. Photography via Pexels (Anna Shvets, Kampus Production, Lisa Fotios). Figures cited are drawn from CNBC on ChatGPT Wallet (November 13, 2026); Unite.AI on Sierra's Personal Agent Protocol 1.0 (November 13, 2026); BikeBiz on Human Forward's AI shopping readiness report (November 13, 2026); PYMNTS, citing the Wall Street Journal, on JPMorgan's agent card block (November 13, 2026); Reuters on Amazon's World Product Graph (November 13, 2026); and Retail Week on Vercel's view of retailer readiness (November 13, 2026).