Agentic Commerce Examples: How AI Agents Shop, Decide, and Pay

Agentic commerce sounds abstract until you watch an agent actually buy something. Then it clicks.

So let me make it concrete. In this post I show how an AI agent shops, decides, and pays in real situations. And I keep two groups apart on purpose: what is live today, at least in an early form, and what is only announced. That gap matters. A press release is not a product. I have learned to read these claims slowly.

The shape of every example

First, the pattern. Every single example below is the same six steps under the surface:

1. Goal — you tell the agent what you want.
2. Discovery — the agent finds products it can read and trust.
3. Selection — it compares the options against your goal.
4. Checkout — it starts the purchase.
5. Authorisation — the system proves you allowed it.
6. Payment — it pays with a safe, limited method.

Keep this in your head. In each example, these six steps are quietly happening.

Example 1: Reorder everyday items

You say: "Order more coffee, same brand as last time, keep it under $30."

The agent finds the product in a store it can read, checks price and stock, adds it to the cart, and pays with a method you approved before. You get a receipt. Done.

This is the most common early example, and for good reason. It is simple, low-risk, and it repeats. Boring is exactly what you want when money moves on its own.

Example 2: Find a specific product under rules

You say: "Find running shoes, size 10, under $120, that can ship to me this week."

Now the agent has to work. It compares many options, checks size, price, and delivery time, throws away anything that breaks your rules, and buys the best match.

This is the part I actually enjoy handing over. It is the slow, tab-opening comparison work that nobody likes doing at 11pm.

Example 3: Manage a subscription

You say: "Cancel the streaming plan I do not use, and find a cheaper one with the same shows."

The agent reviews your subscriptions, cancels one, compares plans, and sets up the new one. This mixes shopping with account actions, which is harder.

It is growing, but it is still limited by what each service lets an agent do. Many companies are not ready to let an outsider cancel things yet.

Example 4: Book travel

You say: "Book a flight to Sydney next Friday, morning if possible, under $400, and a hotel near the centre."

Travel fits agents well. It has clear rules (date, time, price, place) and clear choices. The agent searches, compares, and books inside your limits.

But this one sits mostly in the announced and early group. Real agent bookings are starting, and the amounts are larger, so the limits have to be tighter. I would not hand this one a blank cheque yet.

Example 5: B2B buying

You say, as a business: "Restock our office supplies to the usual levels, but keep this order under the team budget."

A business agent checks stock, builds the order, and buys from an approved supplier — up to a set budget, and maybe only after a manager approves.

B2B is the quiet giant here. The rules are clearer, the buying repeats, and the savings stack up fast. Still early, but moving quickly.

Example 6: Machine-to-machine payments

The setup: one piece of software needs data or compute from another service, and pays a tiny amount for each request. Automatically.

No human clicks at all. One agent pays another. This usually runs on payment rails built for machines, like x402 from Coinbase, often settling in stablecoins.

The numbers here look wild at first. By late April 2026, Coinbase reported roughly 165 million x402 transactions and about $50 million in total volume. But read that carefully: independent analysts found a large share of the early activity was memecoin and test traffic, not real agents buying real services — CoinDesk reporting put genuine daily commerce at only around $28,000. So the rail is real and growing, but do not confuse a big headline number with real adoption. This is a different world from shopping, and it may end up one of the largest parts of agentic commerce.

Live today vs announced

To stay clear-headed, drop any example you read into one of three boxes:

• Live — it works now, even in a small way. Simple reorders and some assistant purchases live here. For scale, OpenAI said ChatGPT reached about 900 million weekly users in February 2026, so even a small share shopping is a lot of people.
• Announced — shown or promised, but few people use it. Much of travel and B2B sits here today.
• Contributed to a standard — a protocol was handed to a standards body. Good for the future, not a working feature now.

Do this one small habit and the loud claims stop fooling you.

What these examples share

From a $20 coffee order to a tiny machine payment, they all need the same three things to be safe:

• Proof that the purchase was allowed.
• Proof of who the agent is.
• A limit on how much it can spend.

This is the part people miss. The hard work in agentic commerce is trust and spending control, not the AI itself. For the basics, see What is agentic commerce?, and for the companies building each piece, see The agentic commerce landscape.

Key takeaway

Agentic commerce is not one thing. It is a pattern — goal, discovery, selection, checkout, authorisation, payment — that repeats across shopping, subscriptions, travel, business buying, and machine-to-machine payments. Some of it is live now. Much of it is still early. The examples that win first are the simple, repeating, low-risk ones.

Sources and further reading

• OpenAI, "Buy it in ChatGPT: Instant Checkout and the Agentic Commerce Protocol" (2025-09-29): https://openai.com/index
• Stripe, "Developing an open standard for agentic commerce" (2025-09-29): https://stripe.com/blog
• Coinbase x402 and the x402 Foundation: https://www.x402.org
• ChatGPT weekly active users (~900M, OpenAI, Feb 2026), reported by TechCrunch: https://techcrunch.com
• x402 real-volume analysis, CoinDesk and Chainalysis (2026): https://www.coindesk.com
• Google Cloud, "Announcing Agent Payments Protocol (AP2)" (2025-09-16): https://cloud.google.com/blog

Note: this space changes quickly, and live/announced status changes with it. Please check the primary sources above before relying on any single detail.