Home / HumaneBytes

AI Can Shop for You. It Can’t Want for You.

ChatGPT Wallet lets AI agents check out on your behalf. That’s real progress. Knowing what you actually want, and when, is still your job.

Outline of a shopping cart with an orange fingerprint inside and an orange speech bubble with a question mark, beside the headline "AI can shop for you. It can't want for you."

On November 13, OpenAI launched ChatGPT Wallet. You save a card once, then your AI agent (OpenAI calls them “dots”) can find what you need and check out for you. For a small group of U.S. ChatGPT Pro subscribers, the distance between “I want this” and “it’s on the way” is now about one conversation long.

It’s genuinely clever, and it was built with care. Before each purchase you review the items, the total and the card, then approve it with Face ID or a fingerprint. The payment network swaps your real card number for a one-time stand-in (a payment token), so neither the agent nor the store ever sees your actual card. Your card rewards and dispute protections still apply.

So this isn’t a “the robots are coming for your credit card” column. It’s about a quieter assumption that tends to ride along with announcements like this one: that because AI can handle the mechanics of buying, it has also handled the decision. Those aren’t the same thing.

The agent executes. You still have to know what you want.

A lot of thinking happens before any purchase, and most of it never gets written down. Should you replace that piece of equipment or nurse it through one more season? Is the cheaper vendor really cheaper once you count the headaches? Is now the right time, or does it just feel urgent because the sale ends tonight?

An agent can compare prices, check delivery dates and finish a checkout faster than any person. What it can’t know is that your priorities shifted two weeks ago when a client contract fell through. Or that you don’t like how a certain supplier treats its workers. Or that you’re simply not ready yet, and that “not ready” is useful information all by itself.

The agent is an excellent errand runner. You still have to send it on the right errand.

Preference is not the same as judgment

Shopping agents work because a lot of buying really is predictable. Purchase history, price sensitivity and brand loyalty are real signals, and a good system can read them well.

Judgment is different. It’s knowing when the pattern should break. The brand you’ve always bought isn’t the right fit anymore. The cheaper option really is the better one this time. Or the smartest purchase is no purchase at all. An AI works inside a model of your past behavior. Good judgment sometimes means deciding the model is wrong.

The “are you sure?” moment is doing real work

Right now, ChatGPT Wallet asks for your fingerprint on every single purchase. That’s a smart way to start. But OpenAI has been clear about where it’s headed: over time, people will be able to hand purchasing to agents with spending limits by amount, category or time period, and pre-approve some purchases so the agent needs less input.

That’s where the interesting questions start. Removing friction is the whole selling point. But friction is also the pause where you catch a mistake, notice you’ve changed your mind, or realize the thing you’re about to buy doesn’t fit the situation anymore. Plenty of us have been saved from a bad decision by the simple act of typing in a card number.

When an agent buys the wrong thing at the wrong time, the agent doesn’t lose any sleep. You’re the one calling the store. The card protections help, but they don’t make the decision for you. Responsibility still lands on a person, and for good reason. People have reputations, relationships and consequences on the line, and that’s a big part of what makes judgment worth trusting.

What to hand off, and what to keep

The useful question isn’t whether to trust AI with purchases. It’s which purchases. A simple way to sort them:

  • Good to delegate: routine, low-stakes buys where your criteria rarely change, like office supplies, printer ink and renewals you’d approve anyway.
  • Keep a human in charge: big commitments, first-time vendors, anything tied to a strategy that’s still changing, and anything where the right answer depends on context the agent can’t see.
  • Set guardrails on day one: when spending limits arrive, start low, review the transaction log weekly, and use a separate card for agent purchases. (It’s also worth knowing what AI companies actually do with your data before you connect much of anything.)

If you sell online, your next customer may be a bot

Here’s the part business owners shouldn’t skip. OpenAI warns that some websites will block agent shoppers without meaning to, and that fraud systems may flag legitimate agent purchases as suspicious, leading to declines or cancellations. It has published a playbook for businesses and is working with payment and fraud-prevention companies to fix that. Some big retailers, Amazon among them, block AI agents outright.

If agents become a real way people shop, the stores they can’t read or check out from simply won’t get the sale. That makes a few old basics newly important: clean product information, a checkout that works without tricks, and a site built to be understood by machines as well as people. That’s the same foundation behind good SEO and AI search visibility, and it’s worth a look if you run an online store. (If you’re on a hosted platform, I wrote about why you don’t fully own your Shopify store.)

ChatGPT Wallet is a real step forward in what AI agents can do, and the convenience is real. But the conversation around tools like this tends to blur the line between carrying out a decision and making one. The agent can close the checkout faster than any of us. Knowing what you actually need, why you need it, and whether now is the right moment is still your job, and it’s worth keeping on purpose rather than giving away by default.

With the holiday shopping season starting, I suspect plenty of dots are about to get a workout. An agent can find the best price on a stand mixer in about four seconds. It still can’t tell you whether you’re actually going to bake.

About the author

Mark Ferguson is CEO and managing partner of Rapid Success Partners, a Plano, Texas business strategy and technology consulting firm. Since 2008 his team has helped companies grow through business consulting, strategic planning, technology consulting, website development and automation and AI.