agentic payment
in English: agentic payments
A payment that you do not initiate manually, but one that an AI agent executes based on your instruction within the limits you set in advance.
In plain terms
In agent-based payment, you do not click the "Pay" button; an AI agent does it for you. You only say what you want and for how much, and it handles the rest: it searches, compares, adds to the cart, and pays.
It is like sending a child to the store with a shopping list and a 5,000-forint note. You told them what they can buy and how much they can spend, but you are not standing at the checkout. The difference is that the agent can check a hundred stores in seconds, and since the store cannot see its face, it must prove in a different way that it is acting on your behalf.
This is new because card payments previously relied on a human on the other side. Fraud filters specifically tried to screen out machine shoppers. Now, rules and technical solutions are being prepared to distinguish the "good" machine from the "bad" one, and to prove afterwards what authorization you gave.
How it works
- You give an instruction: you tell the agent what to buy and under what conditions (for example, maximum price, which store, by when). In more advanced systems, this instruction is recorded in a digitally signed, verifiable format.
- The agent typically does not receive your real card number but a substitute code (token). This is valid only for a specific amount, with a specific merchant, or for a specific period.
- The agent finds the product, assembles the cart, and authenticates itself to the merchant: it indicates that it is an agent, on whose behalf it acts, and what authority it has.
- The merchant, the payment service provider and the bank check whether the purchase falls within the mandate. If it does, the payment usually goes through the same channels as an ordinary card purchase. Some solutions pay not by card but, for example, with a stablecoin (a cryptocurrency pegged to a stable value).
- If you set it this way, or if system rules require it, the agent stops for larger or unusual transactions and requests your approval. The record of the assignment and transaction remains so that responsibility can be determined in case of dispute.
An everyday example
You tell your AI assistant: "If the running shoe I have my eye on drops below 30,000 Hungarian forints in my size, buy it." Two weeks later, you get a notification that the shoe is on the way. The agent watched the price, and when the condition was met, it paid within the given budget.
Who is affected
A common misconception
Limits and risks
- The agent can make mistakes: it may misunderstand your request, choose the wrong size, wrong date, or wrong product, and returning it remains your responsibility.
- A malicious website with hidden instructions may try to trick the agent into buying something else or at a higher price. Therefore, it is advisable to set a low limit and an approval step.
- It is not yet settled who is liable for a faulty purchase: you, the agent's developer, the merchant or the bank. Consumer protection and payment rules only partly answer this so far.
- Multiple competing standards exist, and not every store, bank, or country supports them. You rarely encounter them in Hungarian online shops for now.
Similar, but not the same
- automatic payment (direct debit, subscription) – There, a pre-registered payee receives money, in a fixed amount or the amount on the invoice, while the agent itself decides within the limits what to buy, where and when.
- agentikus kereskedelem – That refers to the entire purchasing process from search to complaint; agentic payment is only the part that handles the money movement.
- mobile wallet (Apple Pay, Google Pay) – With mobile wallets, you typically approve the payment on-site with a fingerprint, facial recognition, or code, whereas with agent-based payment, you grant authorization in advance and the agent initiates the purchase.
Related concepts
AI-agent, agentikus kereskedelem, agentic AI
Agent Payments Protocol, Agentic Commerce Protocol, x402 protokoll, payment mandate
tokenization, Model Context Protocol, inter-agent communication, prompt injection, ember a folyamatban
A brief history
The term became known in 2025. In that year, Visa (Intelligent Commerce) and Mastercard (Agent Pay) announced their agent payment solutions, Google published the Agent Payments Protocol (AP2), and OpenAI and Stripe launched the Agentic Commerce Protocol, on which the first version of in-ChatGPT purchasing was built.
Where things stand
At the end of 2026, the major card networks, payment providers and AI companies are already running live services, mainly in the United States. The standards are still competing with one another, liability and regulatory questions remain open, and only a small share of everyday purchases happens this way.
In the news
- Tencent and Mastercard sign memorandum of understanding on agentic paymentsNov 2, 2026 · Money
- Tencent and Visa to explore AI agent commerce and cross-border paymentsOct 29, 2026 · Money
- Is Nigeria's fraud rulebook ready for AI agents as shoppers?Oct 27, 2026 · Money
- Shopify adopts Visa, Mastercard and American Express agentic payment technology for AI shopping agentsOct 26, 2026 · Money
- Circle, Polygon, Instinct, Hecto and Kakao Pay advance stablecoin payments for AI agentsOct 26, 2026 · Money
- Visa, BILL and Billtrust pilot AI agents for B2B invoice paymentsOct 19, 2026 · Money
- American Express publishes its standard for AI agent paymentsOct 14, 2026 · Money
- Constructor, DaVinci, Gladly and Topsort announce new agentic checkout toolsOct 5, 2026 · Marketing
Also known as
agentic payments · agentic payments · agent payments · AI agent payments · agent-based payment · agentic checkout
Sources
- Agent Payments Protocol (AP2) – official documentation
- Agentic Commerce Protocol (OpenAI and Stripe) – official site
This entry was machine-edited and has undergone machine expert review; the “In the news” section updates automatically from Agent Age stories.