Agentic Checkout
Agentic Checkout refers to the process by which an AI system independently completes a purchase online – without the user confirming every step. The term appears mainly in discussions about AI assistants that shop, book, or subscribe on behalf of humans.
When you buy something online, you click through several steps: add the item to the cart, enter your address, choose a payment method, confirm. In Agentic Checkout, this task is taken over by an AI system, a so-called agent — that is, a program that acts independently on the internet. It decides which product fits, fills out the form, and completes the purchase. The human specifies the goal beforehand, for example: “Order me running shoes under 80 euros in size 43.” The AI handles the rest. That sounds convenient, but it raises serious questions — such as: Who is liable if the AI buys the wrong thing?
Why autonomous payment is changing e-commerce
Today, checkout is the moment where online shops lose the most customers. Studies show that over 70 percent of shopping carts are abandoned before payment — often because of too many clicks or tedious forms. An agent that navigates this path smoothly could significantly reduce this share. For retailers, this is an enormous economic lever.
At the same time, power on the web is shifting. If an AI agent shops for millions of people, it also decides which shops it favors and which it ignores. Retailers then no longer need to convince the human, but the AI. This changes how advertising, recommendations, and pricing work on the internet.
How an AI agent carries out the purchase
An AI agent for this purpose uses three building blocks. First, it understands the user’s request and derives concrete search criteria from it. Second, it navigates through websites — it reads product pages, compares prices, and recognizes where the order form is located. Third, it accesses stored data: delivery address, credit card number, and preferred payment method.
For the last step — the actual payment — to work securely, the agent needs a kind of digital key to the payment method. Credit card companies like Visa and Mastercard are already working on special interfaces that allow agents to pay without knowing the full card number. This limits the risk in case an agent is hacked or misused. Users can also set limits: for example, a maximum of 150 euros per transaction or only at certain shops.
An important difference from passwords saved in the browser: the agent makes decisions on its own. It chooses between providers, compares delivery times, and can independently favor a better offer within the given budget. This is more than autofill — it is an independently acting system.
Agentic Checkout in products and news
The term became widely known mainly through announcements from payment service providers. In early 2025, Visa introduced a program called “Intelligent Commerce” that allows AI agents to pay with Visa cards. Mastercard followed shortly afterward with a similar concept. Major technology companies like OpenAI and Shopify are also working to equip their AI assistants with such purchasing capabilities.
In everyday life, early precursors are already visible: voice assistants like Alexa have been able to trigger simple reorders on Amazon for years. Agentic Checkout goes much further — the agent is meant to act on third-party websites too, not just within a single closed system.
Critics point to risks. If an agent independently enters into contracts — such as an annual subscription for a streaming service — the legal situation is still unresolved. Consumer advocates demand that the human must still explicitly consent before such transactions. This is exactly where the crucial debate lies: how much autonomy should an AI really be allowed to have when spending real money?
