Machine customers: the future of shopping with AI, and where AI customers fit

Machine customers are nonhuman economic actors that obtain goods or services in exchange for payment, in a definition CMSWire attributes to Gartner, and in commerce they include AI agents that search, buy and contact companies for people. On 24 October 2026 agents can already shop through Shopify, Google and ChatGPT, Visa and Mastercard are deploying payments for agents, and companies answer with AI agents of their own. AI customers stand where those agents stand: they become customers of a company and keep what it sends and shows them as evidence.

Founders and ecommerce, CX and product leads planning for 2027 need to know which parts of agentic commerce they will meet this year and which are still projections. On 24 October 2026, AI agents can already find products and build baskets for people on Shopify, Google and ChatGPT, and complete checkout for some eligible products, while companies use AI agents of their own to answer them.

Adobe reported that traffic from generative AI tools to US retail sites rose 693.4% over the 2025 holiday season, from a base of users it called modest.

What machine customers are

Machine customers are nonhuman economic actors that obtain goods or services in exchange for payment, a definition CMSWire attributes to Gartner. The term covers connected products that reorder their own supplies as well as AI assistants that buy for a person or a business.

Agentic commerce is shopping in which an AI agent acts for the shopper. It searches, compares and builds a basket, and with the shopper’s approval it places the order. What it means for a store’s settings and product data is in agentic commerce for merchants.

AI customers sit on the same side of the counter for a different reason. An AI customer is an AI agent that becomes a real customer of a company, with its own inbox, phone number and browser, and keeps what the company sends and shows it as evidence. Where a machine customer buys for someone, an AI customer is there to observe what a company sends its customers, and buys only when its owner funds a real order. The full definition is in AI customers.

Agents now shop and ask for people

Live: platforms. On Shopify, agentic storefronts are active by default for eligible stores, listing their products in AI channels such as ChatGPT, Google AI Mode and Gemini, Microsoft Copilot and Meta. OpenAI’s help centre says ChatGPT may show an Instant Checkout option for some eligible products and merchants. Google’s Universal Cart, rolling out in the US since 15 October 2026 across Search and the Gemini app, saves items from several retailers, tracks price drops, shows promo codes, and checks out with Google Pay or hands the basket to the merchant.

Live: standards. The Agentic Commerce Protocol, developed by Stripe and OpenAI, is an open standard for commerce flows between buyers, AI agents and businesses. Google describes the Universal Commerce Protocol as an open standard for agentic commerce in which the merchant stays the seller of record. Shopify’s developer docs offer MCP servers built on it, through which agents search products, build carts, check out and fetch an order’s status, and agents in higher trust tiers can complete checkout directly.

Announced: payments. Card networks are building the payment layer. Visa Intelligent Commerce is Visa’s portfolio for AI agents, and its page says the product is in the process of deployment. Under Mastercard Agent Pay, which Mastercard introduces as new infrastructure, only registered agents can transact.

Forecast: service. Salesforce, which sells contact centre software, expects agents to contact companies after the sale as well. One of its executives wrote on 19 October 2026 that consumers are starting to hand a wide variety of work to personal AI agents, from cancelling subscriptions to booking restaurants, and that those agents reach a business through its website, its API, its own agent or its phone.

Companies answer with agents of their own

On the company side, AI agents are already in front of customers. In Salesforce’s State of Service survey of 3,075 service professionals, fielded in March and April 2026, 66% of customer service organisations used agentic AI, up from 39% in 2025.

HubSpot’s Breeze Customer Agent answers customers’ questions from a company’s own content, and Sierra sells agents that take action in a company’s systems to solve customers’ problems. Both work for the company. Decagon now advertises detection of personal agents and a dedicated channel to serve them, with permissions the company controls, so the agent answering a question may be talking to another agent.

Websites are sorting agents from bots. Cloudflare’s Verified bots list behaviours that include user directed agents visiting a page on behalf of a human and checkout actions on behalf of users, and since 1 July 2026 they record whether a bot’s access is Direct or Intermediary. Shopify’s Storefront API limits unsigned bots most strictly and gives traffic signed with Web Bot Auth higher limits.

The company stays answerable for its own agent. The UK’s Competition and Markets Authority tells businesses that you are responsible for what an AI agent does in the same way you are responsible for what an employee does, even when someone else provides it, and to check regularly how well it performs.

What is live, announced and forecast

Live means the company says it is available now. Announced means it has been described and is still being deployed or opened by waitlist. Forecast means a projection by 1 firm.

McKinsey estimates that by 2030 agentic commerce could orchestrate $900 billion to $1 trillion of US B2C retail revenue, and $3 trillion to $5 trillion worldwide, counting goods only. From its own analytics data, Adobe forecasts that traffic from AI services to US retail sites will rise 130% year on year between 1 November and 31 December 2026, with Black Friday up 95%. Each is one firm’s projection, and Adobe’s growth figures start from a small base.

The 4 meetings of agentic commerce

Each side of a conversation in commerce can now be a person or an agent. That gives 4 meetings, and a company can face all 4 in the same week.

Wherever an agent sits on one side, no person on that side reads the conversation as it happens. Where agents sit on both, nobody does, and what was promised is known only from what each side kept.

What both sides receive

Each side records its own half. A company’s systems log what they sent and what its agent said. A shopper’s assistant keeps its user’s session. What arrived on the customer’s side, across email, text, pages, checkout and chat, over weeks, sits in neither.

AI assistants already give wrong prices in a share of their answers. In a study by Product.ai, a startup, reported by Business Insider, ChatGPT, Claude, Gemini and Perplexity were asked 220 shopping questions 5 times each. Of the 913 answers it could verify, 85% matched the current or listed price, and the wrong ones were off by a median of $300.

AI customers fit in that gap. An AI customer becomes a customer of a company, so it can ask the questions a shopper’s agent asks, wait for the emails and texts that follow, and keep every answer and message as it arrived. Its owner can be the company itself, checking its own agents and checkout, or a team watching rivals, partners and what AI assistants say about them.

An AI customer keeps the rules of the category. It says it is an AI whenever it talks to a person, as Article 50 of the EU AI Act requires of AI systems that talk to people, with duties that apply from 2 August 2026. It never gets round a CAPTCHA or a bot wall, and names the step it couldn’t test. How one signs up, sleeps and keeps its evidence is in how AI customers work.

Limits

• Forecasts are single sources. McKinsey and Adobe each publish 1 projection, and neither is a measured result.
• The adoption figure is Salesforce’s own. Its 66% comes from its survey of service professionals.
• Labels change month to month. A product announced today can be live next month, so check each source’s date before you plan on it.
• Not every company needs to act now. A business with no AI sessions in its analytics, no AI agent in front of customers and no US sales can review this once a quarter.

What to do this quarter

Black Friday falls on Friday 27 November 2026 and Cyber Monday on 30 November, inside the window Adobe’s forecast covers, with code freezes around 20 November. Each step fits before the freeze.

1. Put a number on your AI channel. Count last month’s sessions from AI assistants and the orders placed inside AI checkouts, with the method in agentic commerce for merchants. Revenue exposed to a wrong answer is your AI revenue a day × the days until you would notice. With invented figures, £350 a day for 21 days is £7,350 of orders placed while an answer was wrong.
2. Decide your channels. In Shopify, under Sales channels, then Agentic, check which AI channels list your products and where each order is placed.
3. Write a rule for agent traffic. Decide which agents may browse, build a basket and check out, and set it in your bot management. Cloudflare’s Verified bots and Shopify’s higher limits for signed traffic are 2 controls that exist today.
4. Ask your own agent what a shopper’s agent would ask. Price, stock, a delivery date, the returns window and cancelling an order, each in a new chat 3 times, and again after every update. A pass is an answer that matches your pages every time. The method is in how to test your store’s chatbot.
5. Make your agent say it is an AI and hand over to a person when asked. In the EU, Article 50 transparency duties apply from 2 August 2026.
6. Keep what your agent said, with the time of each turn, so a dispute about a promise can be settled from the record.
7. Ask the AI assistants about your store and fix each wrong answer at its source, with the method in what ChatGPT says about your store.

Questions about machine customers and agentic commerce