01 — What happened
Three things happened in eleven days, and they only make sense together.
On 9 September, Visa published its Trust Index for agentic commerce — research fielded by Harris Poll among 2,065 US adults on 26–28 May 2026. Seventy-two per cent had used an AI assistant. Twenty-three per cent said they would trust generative AI to execute a payment on its own. Sixty-one per cent named Visa as the brand they would trust to handle an agent-initiated transaction, rising to 68% among 18-to-34s and 71% among frequent AI users.
On 10 September, Ant International, Mastercard and Visa said they would align on a “Know Your Agent” interoperability framework: a common way to tie an AI agent to a validated operator so it can be identified across networks without a separate onboarding for each one. It pulls together three things the same companies had already shipped separately — Visa’s Trusted Agent Protocol, Mastercard’s Verifiable Intent and Ant International’s Agentic Mobile Protocol. The announcement cited a projection of $3–5 trillion of consumer commerce orchestrated by agents by 2030.
On 17 September, Alchemy added Mastercard credentials to its AgentCard product through Mastercard Agent Pay. An agent gets its own email address, phone number, wallet and one-time-use tokenised Mastercard credentials linked to the user’s existing card, with limits on amount, merchant category and geography set by the user and the card issuer.
Now set that against what the platforms where people actually shop decided earlier this year. In March, OpenAI stepped back from completing purchases inside ChatGPT. Its own words: “We’ve found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we’re allowing merchants to use their own checkout experiences while we focus our efforts on product discovery.” Shopify merchants are discoverable across ChatGPT, Microsoft Copilot, AI Mode in Google Search and Gemini — but the purchase completes in an in-app browser or on the merchant’s own store, and the order lands in the merchant’s admin with referral attribution attached. Google’s Universal Commerce Protocol, announced in January, keeps the retailer as the seller of record and asks merchants for catalogue attributes in Merchant Center so their products can be understood in conversation.
02 — Why it matters
Two clocks are running at different speeds.
The payments clock is fast. In under a fortnight the three largest players in agent payments agreed how to identify an agent, and a developer platform shipped a working card credential for one.
The consumer clock is slow. Twenty-three per cent will let the software hold the card — measured among a population where nearly three-quarters already use AI assistants. Familiarity is not the constraint. Handing over the payment is.
Meanwhile the platforms have resolved the ambiguity in your favour, at least for now. The agent’s job is to search, compare and hand over. The handover is the part that is live today, at scale, across four major assistants.
So the question facing a business this quarter is not “should we accept agent payments”. It is narrower and more urgent: when an agent researches this category, does it find us, does it state our facts correctly, and does the person it sends convert?
That is a discovery and conversion problem. Every business already has one. What changed is the reader.
03 — The economics
Being illegible costs nothing and shows up nowhere. No invoice, no error message. An agent assembling a comparison either extracts a specific, current answer from your site or moves to one that provides it. A human fills gaps with goodwill — they ring you, they assume your hours are normal. An agent does neither. It omits you.
The “this is for retailers” reflex is the expensive mistake. A product business with a catalogue can see the path: feed, attributes, stock, price. A service business has no feed and no checkout, concludes there is nothing to do, and misses that the agent-mediated question in its category — who does X in this area, how quickly, on what terms, can they start this month — is precisely the question its website already answers badly.
The work is cheap because it is work you owed anyway. Accurate, structured facts about what you sell, where and on what terms. One source of truth for availability and price. A site that loads. All of it pays for itself in ordinary search and ordinary conversion. Agent readiness is a by-product of that discipline, not a separate budget line.
Building for the checkout is the expensive option, today. Integration effort against protocols that are visibly still converging — three of them being harmonised in September is the definition of unsettled — in service of a transaction fewer than a quarter of people say they want.
The attribution gap is the quiet one. If a growing share of your enquiries arrive already researched and you cannot tell which, you will misread your own marketing and cut the wrong spend. Shopify passes the referral through. Most service businesses log “web enquiry” and lose the signal entirely.
Take an illustrative case: a service business handling 200 enquiries a month, of which a tenth now begin inside an assistant. Those twenty arrive further down the funnel than a cold search visitor — they have already been told you do the thing, in the right area, at roughly the right scale. What loses them is not the absence of an agent checkout. It is a two-day first response, or a form asking for information the agent already collected. The gain sits entirely on your side of the handover — the side you control.
04 — What businesses should do
1. Find out what the assistants currently say about you. Ask three of them, today, the five questions your customers ask most. You will usually find at least one stale fact — old opening hours, a price you no longer charge, a service area you dropped two years ago. Fix it at the source, not in a single page.
2. If you sell products, treat the feed as a revenue asset. Accurate stock, accurate price, machine-readable delivery and returns terms, and the catalogue attributes Google now asks for in Merchant Center. On a mainstream commerce platform most of this is a settings exercise, not a build.
3. If you sell services, build the equivalent of a feed. Pages that state plainly what you do, for whom, in which locations, in what timeframe, with what constraints, and what happens after someone makes contact. Specifics, not brochure copy. “We respond within one working day and quote within three” is machine-readable. “We pride ourselves on service” is not.
4. Instrument the handover before the volume arrives. Add one field to your enquiry form capturing how the person came to you, and watch referrers. If you cannot separate an AI-assisted enquiry from a standard search enquiry within a quarter, you are making budget decisions blind. This is a CRM job, not a growth strategy in itself.
5. Fix response time before anything else. An agent-referred visitor is warm and actively comparing. If you handle several hundred enquiries a month and your first reply takes a day, that is your binding constraint, not your discoverability — and the arithmetic of automating follow-up has not changed.
6. Do not integrate an agent payment protocol yet — unless you are a high-volume retailer whose platform offers it as a toggle, in which case turn it on and measure it. Otherwise, revisit when Know Your Agent is a shipped standard rather than a stated intention, and when the trust number has moved.
05 — Where technology fits
For most businesses reading this, the honest answer is not AI. It is a website content and structure job, plus the discipline of keeping facts in one place.
The integration question is the one that actually bites. If price sits in a quoting tool, availability in a calendar, service area in someone’s head and the website was last edited eighteen months ago, an agent will read whichever version is stalest. That is the familiar cost of disconnected systems, with a new and unforgiving reader. Fixing it means a source of truth and the plumbing to publish from it — the ordinary work of a connected system.
Automation and AI earn their place at the response end, not the checkout end: qualifying and routing the enquiries that arrive, so a warm agent-referred visitor gets an answer in minutes.
And for some businesses the answer is nothing at all. If you win a handful of high-value deals a year through relationships, referrals and tender processes, agentic commerce changes almost nothing for you this year. Correct your basic facts online and get on with the work.
06 — RevenueStack’s take
Our view: the significant development this month is not that an AI agent can now carry a Mastercard credential. It is that the three organisations best placed to make agent payments work spent September agreeing how to identify an agent at all. That tells you the transaction layer is further away than the announcements imply — while the discovery layer is already in production, already shaping what customers believe about you before they arrive, and is almost certainly already sending you enquiries you are not measuring.
The businesses that benefit will not be the ones that integrated a protocol early. They will be the ones whose facts were correct, current and in one place when the machine came to read them. That is not an AI project. It is the same operational discipline that made businesses findable fifteen years ago, applied to a reader with no patience and no goodwill.
Sources
- New Visa Research Finds Consumer Trust is Accelerating the Path to Agentic Commerce — Visa, 2026-09-09
- Visa and Mastercard Team With Ant on Know Your Agent Framework — PYMNTS, 2026-09-10
- Alchemy Unlocks AI Agent Purchases Anywhere Mastercard is Accepted Online via AgentCard — Alchemy (PR Newswire), 2026-09-17
- OpenAI reveals updates to its agentic commerce experience for ChatGPT — Digital Commerce 360, 2026-03-24
- Millions of merchants can sell in AI chats — Shopify, 2026-03-24
- New tech and tools for retailers to succeed in an agentic shopping era — Google, 2026-01-11
