Executive summary: On November 13, 2026, OpenAI launched ChatGPT Wallet. It lets the company's "dots" agents complete purchases on your behalf, built in partnership with American Express, Mastercard, and Visa. Two details change the picture for a small business. First, the agent never touches your real card: after you approve, the payment network issues a payment token — a substitute card number, security code, and expiration date — so neither the agent nor the merchant sees your actual card details. Second, every purchase requires a per-purchase biometric approval (Face ID or a fingerprint), and each approval creates a clear record of what you authorized. The wallet is rolling out gradually, first to US ChatGPT Pro subscribers, 18+, with eligible US-issued cards. For a small business, the transferable capability is not "buy a Pro subscription and let the agent spend." It is the governance pattern: an agent that can now pay is only as safe as the approval gate, the spending limit, and the acceptance check you set. The playbook below is four steps: set the spending envelope, own the approval, read the record, and check the acceptance. EP57 of The AI Guy Podcast walks the same four moves in audio.
Every owner has quietly started delegating the small, boring money. The reorder. The top-up. The bill that is always the same amount. And every one of those used to come back to a decision: did I mean to buy that, at that price, from that vendor? You are not just spending money; you are spending the part of your day you cannot get back, making tiny approvals over and over. An agent that can pay looks like the way out of that loop. But "out" only works if someone still owns the decision. The obstacle is not that the agent cannot pay. It is that a payment agent is only trustworthy while a human is the one who signs, and while there is a written limit on what it may spend. The gate — the approval, the cap, and the acceptance check — is not a limit on the agent. It is what makes you trust it enough to let it prepare the payment in the first place.
Tools & pricing: OpenAI's ChatGPT Wallet (Nov 13, 2026), built with American Express, Mastercard, and Visa. It requires a ChatGPT Pro subscription (OpenAI's paid tier), an eligible US-issued card (Amex, Chase, Mastercard, Visa), and a US billing address. OpenAI has not published a wallet-specific price; the cost is the Pro subscription itself. The capability — an agent that pays with a token and stops for a human approval — is the part to replicate at SMB scale. Competitors are building the same lane: Meta's Business Agent, and the broader agent-payments stack (x402 and related agent-commerce protocols).
Choose the one recurring, low-risk purchase you would be happy to have handled for you, and write its cap. Today, the wallet enforces this by hand: you approve each purchase. Later, OpenAI expects you to be able to delegate by setting parameters — spending limits, purchase categories, or time periods — and pre-approve within them. That automation is an expectation, not a shipped feature. But it will execute whatever limit you have written down. So the envelope you decide this week is the standing order that governs the future. Pick the reorder, the top-up, or the fixed bill. Set the ceiling, the payee, and the frequency. Everything outside that envelope stays with a person.
Tools & pricing: A written rule, not software. The cost is the five minutes it takes to name the purchase and its cap. No subscription required to make the decision.
This is the part that keeps you safe, and the part that separates a tool you trust from a liability you clean up. The agent prepares the purchase; the human signs it, every time. In ChatGPT Wallet, you review the merchant, the items, the total, and the selected card, then approve with a biometric check — Face ID or a fingerprint. OpenAI says each approval creates a clear record of what you authorized. That ten-second review is the control. If you start skipping it because it feels like a chore, you have just handed a stranger a card. The whole safety of this product lives in the part you would be tempted to automate away. Keep the human on the money step, exactly where the money moves.
Tools & pricing: The wallet's own per-purchase approval. No new software. The cost is the few seconds of the review, plus the Pro subscription that gates access.
Every approval creates a record of what you authorized. That record is your audit trail: what the agent did, when, and for how much. Read it on a schedule, the same way you would read a card statement. And remember the caveat OpenAI is explicit about: approval does not confirm the order. The wallet clearing the payment does not mean the merchant placed the order. After the agent pays, you check the merchant's confirmation. That is the gap between "the agent paid" and "the order landed" — and small businesses lose money in exactly that gap. Making the check a habit closes it.
Tools & pricing: The wallet's per-approval record + your own merchant-confirmation habit. No new software. The cost is the habit itself, once a day or once a week.
Not every merchant lets an agent buy from them. Some sites block agent purchases outright, because their fraud systems treat bots as threats. So before you build a workflow where the agent pays for you, confirm that the merchant actually accepts agent payments. For the ones that do not, keep a human on the checkout for that one vendor only. You are not abandoning your card. You are routing the boring, safe, repeatable purchases to the agent, and keeping the human on the ones that do not take it. The pattern is the point. The product is just where you will write it down.
Tools & pricing: A quick check per vendor (can they take agent payments?), not a subscription. No new software. The cost is a one-time mapping of which of your recurring purchases can be delegated.
Four things worth stating plainly. First, this is a limited, gradual, US-only, Pro-subscriber rollout: a limited number of US ChatGPT Pro users, 18+, with eligible US-issued cards and a US billing address. OpenAI itself describes it as "limited availability" while it watches real-world use. Most SMBs are not on the Pro plan at launch, so you are not flipping this on tomorrow. The transferable move is the governance pattern, not the product. Second, it requires a paid ChatGPT Pro subscription; the cost is the subscription, not a wallet fee. Third, approval does not confirm an order, and some merchants will not accept agent purchases — reach depends on the merchant saying yes. Fourth, the "delegate later" capability (spending limits, categories, time windows) is OpenAI's stated expectation, not a shipped feature. None of this is a revenue guarantee; it is a control you set before the automation is real.
The direction is clear. The agent is moving from the text box, to the shelf, to the wallet. Last week it could stand behind your store and answer; now it can pay. And the next step is delegation: you set a spending limit, a category, and a time window, and the agent pays within it without asking you every time. OpenAI says that is the expectation. But the gate that governs it is the one you set today, because the future automation will simply execute the limit you wrote down. The realistic ceiling is honest, not dramatic: this is not a checkout you flip on this week and watch money move itself. What it does is two real things. It proves the agent can pay safely, with your card out of the loop and your signature on every move. And it hands you a record of exactly what you authorized — the closest thing to an audit trail a small business ever gets for the money an agent handles. That is a control asset, and it is a trust asset. The fastest business to set the gate owns the money; the one that owns the gate owns the customer.
Your next move is one approval. Pick the single recurring purchase you would be happy to delegate, and write down its cap, your approver rule, your record check, and your acceptance check. To get the full SMB agentic-AI strategy — including how the finance desk, the phone desk, and the agentic shelf stack with the approval gate — start here:
Get the SMB AI adoption strategy