What happened

On 17 September 2026, developer platform Alchemy announced a partnership with Mastercard: its AgentCard product now plugs into Mastercard Agent Pay, so AI agents can pay with a Mastercard.

AgentCard is a toolkit for people building AI agents. Connect an agent to it and the agent gets its own email address, phone number, stablecoin wallet and one-time-use Mastercard payment credentials, which work as a disposable virtual card linked to the cardholder’s existing Mastercard. With that, the agent can buy on the user’s behalf anywhere Mastercard is accepted online; the examples given are ordering food, booking travel and managing digital subscriptions. Cardholders and issuing banks can set controls on how the agent spends, including purchase limits, merchant categories and where transactions are allowed.

Visa was connected three months earlier. On 18 June, Alchemy launched AgentCard built on Visa Intelligent Commerce, and that announcement put it bluntly: agents can book a vacation, order groceries or renew a subscription on a consumer’s behalf, “without the consumer ever touching a checkout screen.”

Taken together, both major international card networks now let AI agents pay directly. As for where it works, no supported countries have been announced, and neither press release mentions Taiwan.

The merchant end is being wired up too. On 14 September, European payments firm Worldline launched a payment handler for the Universal Commerce Protocol (UCP), an open agentic-commerce standard co-developed by Google and industry partners. Merchants on Worldline configure payments once and can take them across any UCP-connected AI platform. The rails for agent payments are being laid from the card networks right up to the merchant’s till.

Why it matters to you

Until now, being recommended by an AI has been a visibility question. A user sees your name in ChatGPT, Gemini or Perplexity, and then still has to click through, compare prices, read reviews, add to cart and check out. At every one of those steps they might change their mind, and at every one your site, your pricing and your reviews get a chance to win them back.

Once the agent holds the card, that whole path folds into a single step. The user sets a budget and a category, and the agent does the comparing, the choosing and the paying. The moment of deciding who gets the order moves from the person to the agent.

And the agent chooses based on what it understands about each brand: what you sell, what it costs, the specs, how returns work, whether it’s in stock. If that understanding is stale or wrong — last year’s price, a discontinued product still showing, your own site and your marketplace listings telling different stories — you used to lose an impression. From here, you lose an order. You won’t see it happen, either. When an agent buys from someone else, no abandoned cart turns up in your dashboard for you to chase.

What to do about it

There’s no word yet on when this reaches Taiwan, but which side moves fast and which moves slowly is already clear. Payments move fast: Alchemy went from Visa to Mastercard in three months. What moves slowly is an AI’s picture of a brand, pieced together from scattered descriptions across the web. Correct a wrong price and you still have to wait for each AI to read the correction before it counts. That takes time, and there’s no button that makes it happen. Start tidying up on the day agents can check out for Taiwanese shoppers, and you walk in carrying an old impression.

What wears you down is the detail. An agent comparing options compares specific facts, one at a time: this price, this spec, whether it’s in stock, how returns work. Those facts are scattered across your own site, your product pages on each marketplace and comparison sites, often updated by different people at different speeds. The price changes on your site while the marketplace still shows the old one; a product is discontinued and pulled from one channel while another still lists it as in stock. In the few seconds an agent spends comparing, one stale entry is enough to get you dropped. So someone has to line that data up, channel by channel and item by item, and then do it again every time a price changes, a season turns or a returns policy is adjusted.

When AI talks about your prices, specs and returns terms today, which version is it working from? Book a demo and we’ll lay out what it’s saying right now.

Further reading