The customer that's a machine: agentic commerce without jargon
A new customer is being born: the AI agent that researches, chooses, and buys on behalf of your real customer. The ACP (OpenAI/Stripe) and UCP/AP2 (Google/Shopify/Mastercard, NRF 2026) protocols are already standing. This lesson shows you what changes, why the machine-readable catalog becomes a revenue lever, and the fatal mistake: outsourcing the customer relationship to a platform.
Picture a Saturday morning. Before, your customer would grab their phone, type what they wanted, look at three sites, compare prices, read a review, hesitate, and maybe buy. You had several moments to show up: the ad, the page, the pretty photo, the button. Now imagine they just tell their AI assistant: "get me the best running shoes under three hundred bucks, delivered by Friday". And that's it. The agent goes out, researches, compares, chooses, and buys. Your customer never even saw your store. The question this lesson answers isn't "how do I advertise better?". It's: when the one deciding the purchase is a machine, is your product written in a way that machine can read, trust, and recommend?
A CFO tells the company's copilot: "renew the three software subscriptions expiring this month for the best cost-benefit." The agent evaluates, chooses, and closes payment via protocol, without anyone visiting a single vendor site. Whoever wins the renewal isn't whoever had the best banner; it's whoever had the clearest product sheet, with price, terms, and a comparison the machine could read. B2B purchasing is turning agentic too, and your commercial proposal needs to be legible to a reader that has no eyes.
A law firm uses an assistant to hire a deadline-management tool. Instead of the partner browsing and comparing, the agent reads the sheets, cross-references requirements (LGPD, exit clause, Portuguese-language support), and recommends. The firm that wrote those terms in a structured, verifiable way shows up in the answer; the one that buried everything in a pretty marketing PDF disappears. To the machine, whatever isn't legible and structured simply doesn't exist when it's time to recommend.
You run e-commerce for a cosmetics brand. Half your energy goes into photos, videos, ads, creative, things made for the human eye. Then agentic commerce arrives: the customer has their agent buy "a good moisturizer for sensitive skin, unscented, under eighty bucks". The agent doesn't look at your beautiful photo; it reads your product's attributes. If your sheet says "unscented: yes", "skin type: sensitive", "price: 79", you show up in the answer. If that's only in the image or in poetic copy, you're left out. The storefront changed readers, and most brands are still decorating the storefront for the wrong reader.
Picture this: a manager tells the HR copilot "bring me the top three candidates for this mid-level analyst role, with SQL experience and availability for hybrid work." The agent scans the database, reads the profiles, and recommends. The candidate who described their experience in a structured way (skill: SQL, seniority: mid-level, model: hybrid) shows up on the list; the one who buried everything in a pretty PDF résumé, with the skill diluted in a poetic paragraph, disappears. To the machine, talent that isn't legible and structured simply doesn't make the shortlist. Your candidate's storefront changed readers.
You lead product, and your team uses an agent to choose which analytics tool to integrate into the roadmap. Instead of the PM browsing and comparing, the agent reads each solution's sheets, cross-references them with the PRD's requirements (custom event support, mobile SDK, price per event), and recommends. The tool that published those attributes in a structured way makes the prioritization; the one that left the specs in a marketing deck gets left out. When the one comparing is a machine, what matters isn't the product's nice narrative, it's the attribute it can read and trust.
Picture a buyer telling their company's copilot "close with the supplier who delivers within five days, with an invoice, and the best per-unit price." The agent won't fall for your polished prospecting or your gorgeous PowerPoint proposal; it reads the attributes: lead time, terms, price, condition. Whoever structured the proposal in a way the machine can read enters the agent's pipeline and closes; whoever sent everything in a beautifully designed PDF, with numbers scattered through the text, isn't even considered. Your forecast, going forward, depends on your offer being legible to a buyer that has no eyes.
You run operations, and your company starts using an agent to pick a carrier for every order. The agent reads the sheets: delivery SLA, region coverage, damage rate, cost per route, and recommends on the spot. The carrier that published those indicators in a structured way enters the automatic selection; the one that keeps everything in a descriptive report, with no legible field, gets left out of the flow. For the machine orchestrating logistics, the supplier that isn't structured simply doesn't exist at dispatch time. The storefront turned into an attribute sheet.
Picture a compliance team using an agent to hire a risk-monitoring tool. The agent cross-references requirements (LGPD adherence, audit trail, log retention, certification) against vendors' sheets and recommends. Whoever described those controls in a structured, verifiable way shows up in the answer; whoever buried it all in a pretty institutional brochure disappears from the comparison. For the machine assessing compliance, a control that isn't legible and auditable is as good as nonexistent. What protects you is the attribute written to be read, not the presentation made to be admired.
Picture the engineering team's agent choosing which observability service to adopt: "find me one with a Node SDK, thirty-day retention, Slack alerts, and the best cost per GB." The agent reads each vendor's structured docs, compares, and recommends, without anyone opening a landing page. Whoever exposed those attributes in a legible way (compatibility, limits, price) makes the decision; whoever left everything in a poetic blog post gets left out of the deploy. When the one deciding the integration is a machine, what matters is the spec sheet it reads, not the pretty site no agent ever opens.
You work in UX, and your team uses an agent to choose a usability testing tool to adopt in the workflow. The agent reads the sheets, cross-references requirements (session recording, prototype support, WCAG accessibility, Figma integration), and recommends. The tool that structured those attributes gets chosen; the one that hid everything behind a gorgeous site with no legible field disappears from the comparison. There's a nice irony here: even the design tool is now chosen by a reader with no eyes, one that doesn't care about aesthetics, only about the attribute it can read and trust.
Picture a strategy director telling the company's copilot "find me three market-data providers with coverage of my sector, monthly updates, and an API to plug into our OKR dashboard." The agent reads each provider's spec sheet, cross-references the requirements, and recommends, without anyone opening an institutional website. Whoever published coverage, frequency, and format in a structured way makes the comparison; whoever only has a pretty PDF report, with no legible field, gets left out. When the one deciding the contract is a machine, what matters is the attribute it can read, not the most elegant sales deck.
Let me introduce you to a new customer who just walked into your store, one you probably haven't noticed yet. It has no eyes, doesn't get emotional about your photo, doesn't fall for your headline. It's a machine, an AI agent, and it's buying on behalf of a real person. This isn't conference-stage future talk. In September 2025, OpenAI and Stripe launched a standard for agents to buy inside ChatGPT, and Google, Shopify, Visa, and Mastercard launched their own. At NRF 2026 (the world's biggest retail trade show), this became the central topic. This lesson is here to help you understand, without jargon, what changes when the one pressing the buy button is no longer a human, and what mistake could turn you into an anonymous supplier of a storefront that isn't yours.
The core idea of this lesson. A new layer of purchasing is being born: agentic commerce, where an AI agent researches, chooses, and closes the purchase on behalf of your customer. There are already rails laid down for this (the ACP, UCP, and AP2 protocols, which you'll understand in plain language). This changes two things in your marketing. First: your machine-readable catalog (price, attributes, stock, terms, all structured) stops being a technical detail and becomes your main storefront and a direct revenue lever. Second, and this is the warning I don't want you to forget: don't outsource the relationship with your customer to a platform. If the only place your customer finds you is inside another company's agent, you don't have a customer, you have a middleman who can swap you out any time.
01What agentic commerce is, in plain language
Let's get to the concept without beating around the bush. Agentic commerce is when your customer's AI agent makes the purchase for them. The human gives the intent ("I want this, with these conditions"), and the agent executes the entire journey: searches, compares, chooses, and pays. The customer becomes the boss who delegates; the agent becomes the assistant who goes to the market.
Notice how different this is from the search you already know. In traditional search, AI showed you an answer and you still clicked, entered the site, decided. (That's the lesson about being cited by AI, the so-called GEO, which you saw earlier in this track.) Agentic commerce goes one step further: the agent doesn't just recommend, it acts, it buys. The click disappears, the site visit disappears, the storefront as you knew it disappears. The decision happens inside a conversation with AI, in a place your Google Analytics doesn't even see.
And don't think this is just sneaker retail. It applies to software (renewing a subscription), to services (hiring a tool), to B2B (a copilot closing a company's recurring purchase). Wherever there's a repeatable, comparable purchase decision, the agent shows up. Your marketing work, going forward, splits into two audiences: the human (who still gets emotional) and the machine (which reads attributes). Whoever only talks to the human is losing the half that's growing.
02The protocols, no acronym memorization needed: ACP, UCP, and AP2
Now the names you'll hear in the meeting, translated. You don't need to memorize acronyms, you need to understand what each one is for, because they're the rails this train is going to run on. Think of them as the traffic rules that make the agent, the store, and the bank understand each other at purchase time.
The first is the ACP, Agentic Commerce Protocol, from OpenAI with Stripe (September 2025). It's the standard that makes the purchase happen inside ChatGPT: the agent builds the cart, the store confirms the order, and payment closes, without the customer leaving the conversation. It's "buying inside ChatGPT". Big names are already on board: Etsy, Shopify, and Salesforce. It's worth the honesty the news itself brought: OpenAI's first attempt stumbled and they're on their second round. In other words, it's being born, not finished. But it's being born for real.
The second is the UCP, Universal Commerce Protocol, from Google (with Shopify, Walmart, Target, Etsy). This one is more ambitious: it doesn't just cover "closing the purchase", it covers the entire journey, discovery, purchase, and post-sale (returns, tracking, support). Think of ACP as the store's checkout and UCP as the entire store, from the aisle to the returns counter.
The third is the AP2, Agent Payments Protocol, also from Google, with over sixty partners including Mastercard, PayPal, and Amex. This one solves the trickiest part: how the agent pays securely on behalf of a person without spreading their card details around. It uses a kind of "stamped authorization" (a verifiable mandate) that proves the customer really did grant permission for that purchase. It's what gives some peace of mind to the idea of a machine spending your money.
What matters to you, marketing leader, isn't implementing any of this (that's the job of whoever handles checkout). It's understanding that these rails exist, that big tech and card networks are betting on them, and that they turn "the agent recommended" into "the agent bought". When the rail is ready, the train passes. The question is whether your product will be at the station, legible, or sitting in a warehouse somewhere, hidden.
03The machine-readable catalog becomes a revenue lever
Here's the practical turn of this lesson, and it's the best news in it. When the buyer is a machine, what it reads isn't your photo or your poetic copy. It reads your product's attributes: name, price, stock, color, size, "unscented: yes", "recommended for: sensitive skin", delivery time, return policy. All of it structured, written to be read, not to be admired.
In the market they call it a golden record, the product sheet with nearly all attributes filled in and organized. And there's a logic worth printing out and taping to the wall: stores with well-filled catalogs show up far more in agent recommendations than stores with sparse, gappy sheets. Think about what this means. It's not a cosmetic improvement; it's revenue. Every attribute missing from your sheet is a search where the agent doesn't consider you, because it has no way of knowing you meet that criterion.
This is a cousin of what you already saw about feeding AI the right context (the context lesson and the RAG one, back at the course's foundation). It's the same principle applied down to your product: AI only recommends with confidence what it can read and understand clearly. The difference is that now it's not just for it to cite you in an answer, it's for it to buy from you.
And notice the priority inversion this causes. For years, the product catalog was the boring task, handed off to the intern, "just fill it in with whatever". The pretty photo took the whole budget. Going forward, a clean, complete catalog is one of the biggest revenue levers you have, and it costs much less than a campaign. It's the unglamorous work that separates whoever sells to the agent from whoever gets left out. The storefront changed readers; where you invest your attention needs to change with it.
04The fatal mistake: outsourcing the customer relationship
Now the part I can't let you forget, because it's where the long-term risk lives. Agentic commerce is a wonderful convenience, and precisely because of that it has a trap: it puts a platform between you and your customer. And if you're not careful, that platform becomes the owner of the relationship, and you become the anonymous supplier down at the bottom.
Think about what happens. The customer talks to the agent, not to you. The agent knows their taste, their history, knows they prefer fast delivery and hate scented products. That knowledge, that relationship, stays entirely with the agent's platform, not with your brand. Next time, if the agent decides to recommend the competitor (because they paid more, because the algorithm changed, because they struck a deal), your customer won't even notice you ever existed. You were disposable, because the customer's loyalty is with the agent, not your brand.
It's the same mistake as whoever built their entire business on top of a social network and woke up one day with their reach cut in half. The platform changes the rule, and whoever had no direct relationship with the customer is left holding the bag. In agentic commerce, the squeeze is even tighter, because the platform doesn't just mess with your reach, it messes with who gets recommended to buy.
So the golden rule, and it's strategic, not technical: be present in the agents, but never depend only on them. Use agentic commerce as one more channel (and yes, prepare your catalog for it), but keep building a direct relationship with your customer: your list, your contact, your reason for them to look for you by name, not just accept whatever the agent suggested. Whoever is cited and bought by AI wins the present; whoever also has a direct relationship with the customer wins the future. Both, not just one.
05The map: what you do with this now
Put the three pieces together and you have your move. One: agentic commerce is real, it has rails laid down (ACP, UCP, AP2) and it's coming in through retail and B2B. Two: your new storefront is the machine-readable catalog, and filling it out well is a direct revenue lever, not an intern's task. Three: never hand the customer relationship over to the agent's platform on a silver platter; be there, but have your own direct channel.
The rest of this marketing module picks up each of these pieces and turns it into hands-on work: how to make your brand and product legible to AI, how to measure whether you're showing up in agents' answers and purchases, and how to keep your voice and customer relationship without becoming hostage to someone else's storefront. This lesson was the framing; now we install it.
To take with you: agentic commerce is being born, where the AI agent buys on behalf of your customer, over rails already standing (ACP from OpenAI/Stripe, UCP and AP2 from Google with Mastercard and retail, front and center at NRF 2026). This turns your machine-readable catalog (complete, structured attributes, the golden record) into your main storefront and a direct revenue lever, well-filled stores show up far more in recommendations. And it imposes a strategic caution: be present in the agents, but never outsource the customer relationship to the platform, or you'll become an anonymous supplier of a storefront that isn't yours. Fair enough? Next up.
Do it now
Your mission for this lesson is to run the audit of your storefront for the machine, in about fifteen minutes, taking ONE real product or service of yours (your real task or another).
PART A · THE LEGIBLE CATALOG (the revenue lever) Pick a product and imagine your customer's AI agent reading its sheet. List what's actually written in a structured way (not in the photo, not in poetic copy, but in a field a machine reads): price, stock, attributes (color, size, "unscented", "recommended for"), delivery time, return policy. Now honestly mark how many of these fields are complete and how many are empty or only in the image. Ask the hard question: if the agent compared my product with the competitor's right now, would it have enough information to choose me, or would I disappear for lack of attributes?
PART B · THE DIRECT RELATIONSHIP (the insurance against the platform) Answer each in one sentence: today, if an AI agent stopped recommending me tomorrow, would I be able to talk directly to my customers? Do I have a list, contact, community, some reason for them to look for me by name? Or is the only bridge between me and the customer a platform I don't control?
At the end, write your TWO next actions: (1) which catalog attribute you'll fill in first, because it's what most keeps you out of recommendations today; (2) which direct-relationship step you'll take so you don't depend only on the agent. You don't need to understand any protocol to do this audit. You need to look at your storefront through a machine's eyes, and at your brand through the eyes of someone who doesn't want to be disposable.
Where the protocols come from and why this exploded in late 2025
The language you'll hear from the technical team is full of acronyms, so here's a quick map of the real sources, to recognize them when they show up in a proposal. ACP (Agentic Commerce Protocol) was announced by OpenAI in "Buy it in ChatGPT" and detailed by Stripe as an open standard, in September 2025: it's the rail for "buying inside ChatGPT", with Etsy, Shopify, and Salesforce among the first. UCP (Universal Commerce Protocol) is Google's response, documented on their developer blog (with Shopify, Walmart, Target, and Etsy), and covers the whole journey (discovery, purchase, post-sale), compatible with other agent standards. AP2 (Agent Payments Protocol), also from Google Cloud, solves delegated payment with over sixty partners, including Mastercard, PayPal, and Amex, using verifiable "mandates" (the stamped authorization that proves the customer consented). It's worth the realism CNBC brought in March 2026: OpenAI's first agentic purchase attempt stumbled and they went to a second round. Translation: it's an enormous bet by big tech and card networks, still in a maturing phase. You don't need to implement any of this. You need to recognize the names, understand that the rails are being built for real, and prepare your catalog and customer relationship before the train passes full.
Practice
1. What best describes agentic commerce?
2. Why did the machine-readable catalog (complete, structured attributes) become a revenue lever, and not just a technical task?
3. Your brand started selling well through a major AI platform's agent. What's the strategic risk this lesson tells you to watch for?
Let's close the point together, because it changes your marketing game. A customer that's a machine has entered the scene: the AI agent that researches, chooses, and buys on behalf of your real customer, over rails the world's biggest companies are already building (ACP, UCP, AP2). You don't need to implement any protocol, but you need to do two things most haven't done yet. Prepare your storefront for this new reader, by filling out the catalog legibly and completely, because that's where real revenue lives, not a technical detail. And protect your direct relationship with the customer, using agents as one more channel, never the only one, so you don't wake up disposable inside a storefront that isn't yours. Whoever understands this now prepares the house while the train is still arriving. Whoever ignores it will find out, too late, that the customer stopped seeing them, and didn't even notice it stopped. Let's move to the next one.
For the board
On the new customerit does not fall for your headline. Someone delegated the intent to it, and it runs the whole journey.
On the cataloguewhen the buyer is a machine, the readable attribute is the shop window. It is the unglamorous work that separates who sells from who disappears.
On dependencyloyalty stays with the agent, not with your brand. Whoever also owns a direct channel owns the future.
Thanks for the feedback. It helps sharpen the next lesson.