Updated November 17, 2026.
Agentic commerce is commerce in which AI agents discover products, compare options, and complete parts of a purchase on behalf of a person or a business, within limits the user sets. Agentic payments are the payment step of that flow: the agent starts the transaction with a tokenized credential instead of the user's real card or account details, and the user's approval is recorded as proof of what they authorized.
By November 2026, the infrastructure for this is live but early. On November 13, 2026, OpenAI launched ChatGPT Wallet, built with American Express, Chase, Mastercard, and Visa and available first to ChatGPT Pro subscribers in the US. Mastercard says all its issuers in Europe are enabled at a network level for Agent Pay (June 2, 2026), and Visa is piloting Intelligent Commerce Connect with select partners (April 8, 2026). The first attempt at checkout inside a chat did not go as planned: in March 2026, OpenAI let merchants return to their own checkout and refocused on product discovery.
For fintech companies, this changes how payments are authorized, how intent is verified, how fraud is detected, and how checkout is built. Below: how an agentic payment works, which protocols matter, what adoption looks like in numbers, and what merchants, PSPs, and banks need to build.
Agentic commerce is a model where AI agents act on behalf of a person or business during the buying process. Visa describes it as commerce where "AI agents help consumers or businesses discover products, make decisions and complete parts of the purchasing journey" (Visa Intelligent Commerce). The agent can search for products, compare options, check prices, apply rules, prepare checkout, and in some cases initiate or complete a payment after the user gives permission.
The main difference from traditional e-commerce is the role of the interface. In traditional commerce, the user visits a website or app, browses manually, and clicks through checkout. In agentic commerce, the AI assistant becomes the first layer of interaction. It may recommend products, filter suppliers, compare terms, and send purchase details to the merchant through an API.
For fintech, the key question is not whether agents can recommend products. It is how money moves safely when an AI agent is involved in the transaction. This requires clear authorization, payment credentials, transaction limits, consent, fraud controls, audit trails, and dispute processes.
Agentic payments are transactions that an AI agent initiates on a user's behalf, using credentials and permissions the user has delegated. The money still moves over existing rails, such as cards, bank transfers, or stablecoins. What changes is who presses "buy", which credential the merchant receives, and what evidence exists of the user's consent. For a wider look at where agents already sit in payment flows, see how AI agents are changing digital payments.
The difference is easiest to see next to a regular online card payment.
| Regular online card payment | Agentic payment | |
|---|---|---|
| Who starts the purchase | The cardholder, on the merchant's checkout page | An AI agent, after the user sets a task or approves a cart |
| What the merchant receives | A card number or a device wallet token | A token issued for the agent's purchase by the network or the platform, not the real card number |
| How the user confirms | 3-D Secure, a wallet prompt, or no extra step | Biometric or passkey approval of the specific purchase, or limits set in advance by amount, merchant, category, or time |
| Proof of intent | Session and device data | A recorded approval or a cryptographically signed mandate (Google AP2) |
| How the merchant recognizes the buyer | Browser session, cookies, fraud signals | Agent verification, such as Visa Trusted Agent Protocol or signed requests (Web Bot Auth) |
| Disputes | Standard card dispute rules | Card protections still apply in card-based wallets such as ChatGPT Wallet; rules for fully delegated purchases are still forming |
Most agentic payment schemes follow the same sequence: the user delegates, the agent prepares the purchase, the user or a pre-set rule approves it, the network issues a token, and the merchant charges it. OpenAI's description of ChatGPT Wallet shows each step.
The user adds a card once. In OpenAI's words, "Users add a card once, then approve each purchase without re-entering their payment details."
The agent prepares the purchase. It finds the product and builds the cart on the merchant's website.
The user reviews and approves. The user checks the items, the total, and the selected card, then confirms with a security check such as Face ID or a fingerprint scan. Each approval creates a record of the purchase the user authorized.
The network issues a token. After approval, "the payment network provides a payment token—a substitute for the user's actual card details—with a card number, security code, and expiration date."
The merchant charges it as a normal card payment. The business receives the token, not the real card number, and keeps control of pricing, fulfillment, returns, and customer service. The user keeps card rewards and protections for eligible disputes.
Two details matter for payment teams. First, approval is per purchase today. OpenAI expects users to delegate purchases entirely later, by setting "spending limits, purchase categories, or time periods." Second, merchants have to let agents through. OpenAI's guidance for businesses asks them to configure their CDN or firewall to recognize signed requests using Web Bot Auth, and to work with payment and risk providers to reduce unnecessary declines.
Fintech companies operate at the point where customer experience, payments, risk, identity, and regulation meet. Agentic commerce affects all of these layers at once.
The Agentic Commerce Protocol (ACP), developed by OpenAI and Stripe, started as the rail for checkout inside ChatGPT. Since March 24, 2026, OpenAI has used it as "the connective layer between merchants and users throughout discovery": merchants share product feeds and promotions through ACP, while checkout can stay on the merchant's side. In November 2026, the payment step moved to network tokens with ChatGPT Wallet.
This means fintech infrastructure has to evolve. Payment systems were built around human clicks, browser sessions, card entry, and checkout pages. Agentic commerce requires API-first checkout, tokenized payment credentials, delegated permissions, real-time risk scoring, and machine-readable product and pricing data.
In a normal checkout flow, the customer controls every visible step. They choose the product, enter payment details, confirm the order, and complete the purchase.
In an agentic flow, the user may give an AI agent a task such as "find the best business travel option under this budget" or "reorder office supplies when prices drop." The agent then compares options, checks conditions, prepares the cart, and requests payment authorization or acts within pre-approved limits.
Google's Agent Payments Protocol was created to address exactly this problem. Google explains that today's payment systems usually assume a human is directly clicking "buy" on a trusted surface, while autonomous agents break that assumption. AP2 focuses on authorization, authenticity, and accountability for agent-led payments.
For businesses, this changes the payment journey in several ways:
Discovery becomes AI-led. Customers may discover products through AI assistants instead of search engines, marketplaces, or ads.
Checkout becomes API-driven. The checkout process has to work through structured data and secure payment flows, not only website forms.
Authorization becomes delegated. The user may approve specific rules, limits, merchants, product categories, or time windows.
Fraud monitoring becomes more complex. Businesses need to detect suspicious behavior from both humans and agents.
Customer experience becomes agent-readable. Product data, pricing, availability, shipping, loyalty benefits, and return rules need to be understandable for AI systems.
Agentic commerce is still developing, but several use cases are already clear for businesses and enterprises.
AI agents can help users compare products, apply loyalty benefits, choose a payment method, and pay with approval. ChatGPT Wallet is the first large consumer example: the agent pays with a network token after the user approves the purchase with a biometric check, and merchants keep order processing and fulfillment in their existing systems.
Digital wallets can become the control layer for agentic commerce. A wallet can manage payment methods, spending limits, approvals, merchant restrictions, recurring payments, and transaction history.
For neobanks, this creates an opportunity to offer AI-powered financial assistants that help users manage purchases, subscriptions, bills, travel, budgets, and savings decisions.
Payment service providers can support agent-ready checkout, tokenized credentials, merchant validation, risk scoring, and payment orchestration.
This is especially important because businesses may not want to integrate separately with every AI assistant or commerce platform. Protocols such as ACP, AP2, and UCP aim to reduce fragmentation by creating common rules for agent-to-business transactions.
In enterprise procurement, AI agents can compare suppliers, validate invoices, check approval rules, confirm budgets, and prepare payments.
This can reduce manual work in finance teams, especially for recurring purchases, travel, software subscriptions, vendor payments, and invoice review.
Agentic commerce can make embedded finance more intelligent. Instead of offering payments, credit, or insurance at a fixed checkout step, businesses can provide financial options through AI agents that understand user intent and context.
For example, an agent could compare payment terms, select a card with the best rewards, recommend a financing option, or route a transaction through the most efficient payment rail.
Google notes that AP2 supports payment types "from credit and debit cards to stablecoins and real-time bank transfers." Machine-to-machine payments now have their own protocols as well: x402, contributed by Coinbase to the x402 Foundation at the Linux Foundation, and the Machine Payments Protocol from Stripe and Tempo both accept stablecoins, and Mastercard Agent Pay for Machines works across cards, accounts, and stablecoins.
For fintech companies working with crypto or Web3 infrastructure, this creates new demand for wallets, stablecoin payment rails, programmable payment logic, and transaction monitoring.
ilink can provide scalable fintech solutions for your business.

Agentic commerce runs on several protocols, and each solves a different part of the problem: product discovery and checkout, proof of user intent, agent identity, payment credentials, or settlement. None of them covers the whole flow, and card networks are building on-ramps that accept several at once. The table shows where each stood in November 2026.
| Protocol | Who and when | What it solves | Payment methods | Status, November 2026 |
|---|---|---|---|---|
| Agentic Commerce Protocol (ACP) | OpenAI and Stripe, 2025 | Product feeds and discovery; originally checkout in ChatGPT | Merchant's own payment stack | Extended to discovery on March 24, 2026; merchants can use their own checkout |
| Agent Payments Protocol (AP2) | Google, September 16, 2025 | Proof of user intent through signed mandates | Cards, stablecoins, real-time bank transfers | Open protocol; more than 60 organizations involved at launch |
| Universal Commerce Protocol (UCP) | Google with Shopify, Etsy, Wayfair, Target, Walmart; January 11, 2026 | Discovery, cart, checkout, and orders between AI surfaces and merchants | Pluggable payment handlers; works with AP2 | Open-source; endorsed by more than 20 partners |
| Visa Trusted Agent Protocol and Intelligent Commerce Connect | Visa; Connect announced April 8, 2026 | Agent verification and one on-ramp for payments started through TAP, MPP, ACP, and UCP | Visa and non-Visa cards | Connect in pilot with select partners |
| Mastercard Agent Pay | Mastercard | Tokenization, authentication, consumer consent | Mastercard cards | All issuers in Europe enabled at a network level (June 2, 2026) |
| Mastercard Agent Pay for Machines | Mastercard, June 10, 2026 | High-volume, low-value payments by autonomous agents | Cards, accounts, stablecoins | Launched with more than 30 technology partners |
| Machine Payments Protocol (MPP) | Stripe and Tempo, March 18, 2026 | Agents paying services programmatically, including microtransactions | Stablecoins; cards and buy now, pay later via Shared Payment Tokens | Open standard; accepted on Stripe |
| x402 | Coinbase; x402 Foundation at the Linux Foundation, July 14, 2026 | Payments over HTTP for agents, APIs, and apps | From cards to stablecoins | Vendor-neutral foundation with 40 members at operational launch |
ACP was developed by OpenAI and Stripe to let AI agents, users, and businesses work together on purchases. Since March 2026, OpenAI uses it mainly for discovery: "Through ACP, merchants share product feeds and promotions so their catalogs are fully represented in ChatGPT" (OpenAI).
Google's AP2 is an open protocol for agent-led payments across platforms. It uses mandates, which Google describes as "tamper-proof, cryptographically-signed digital contracts that serve as verifiable proof of a user's instructions." AP2 supports real-time purchases where a human is present and delegated tasks where the human is not present at checkout.
UCP covers the shopping side: discovery, cart, checkout, and orders. Businesses publish their capabilities in a manifest at /.well-known/ucp, connect through APIs, Agent2Agent (A2A), or the Model Context Protocol (MCP), and remain the merchant of record. Google says UCP "is compatible with Agent Payments Protocol (AP2) to provide secure agentic payments support" (Google Developers Blog).
The card networks build on tools they already have: tokenization, authentication, and consumer consent. Visa's Trusted Agent Protocol lets merchants verify that an agent is legitimate, and Intelligent Commerce Connect is a network-, protocol-, and token-vault-agnostic on-ramp for payments started through four agent protocols. Mastercard wrote on June 2, 2026, that "all Mastercard issuers in Europe are enabled at a network level for Agent Pay," with live transactions authenticated by passkeys.
Not every agent payment is a consumer purchase. Agents also pay for API calls, data, and compute, often in amounts too small for card economics. Three schemes target this: x402, which uses HTTP to request and settle payment and moved to vendor-neutral governance at the Linux Foundation; the Machine Payments Protocol from Stripe and Tempo, which enables "microtransactions, recurring payments, and more"; and Mastercard Agent Pay for Machines, launched on June 10, 2026, for high-volume, low-value payments.
Adoption forecasts are large, but live volume in 2026 is still small and mostly in pilots. Three figures frame the market:
McKinsey (October 2025): by 2030, AI agents could orchestrate up to $1 trillion in US B2C retail revenue and $3 trillion to $5 trillion globally (McKinsey).
Accenture (May 27, 2026): "In 2030, we estimate that more than 30% of online commerce could run through AI agents, representing close to $3.1 trillion in transactions" (Accenture).
OpenAI (November 13, 2026): 1.2 billion people use ChatGPT each week, and many of them use it to research products, compare options, and decide what to buy.
The challenges are visible in what happened during the year:
Checkout inside the AI did not work as planned at first. OpenAI wrote on March 24, 2026: "We've found that the initial version of Instant Checkout did not offer the level of flexibility that we aspire to provide, so we're allowing merchants to use their own checkout experiences while we focus our efforts on product discovery."
Network programs are still in early stages. Visa Intelligent Commerce Connect is "currently in pilot with select partners," and Mastercard describes its European transactions as "controlled, live agentic transactions."
Merchant systems block agents. OpenAI notes that "some sites may unintentionally block purchases through ChatGPT Wallet or other agentic payment products."
Standards are fragmented. The eight protocols above overlap, which is why Visa built an on-ramp that accepts four of them.
For financial institutions, the practical conclusion is to build the control layer now: tokens, limits, consent records, and dispute handling. Waiting for one standard to win is riskier, because the networks themselves are designed to support several.
Agentic commerce can create value for fintech companies, merchants, PSPs, banks, and enterprises if the infrastructure is designed correctly.
Faster checkout. Agents move users from intent to purchase faster when they have accurate product data, checkout APIs, and availability.
A new acquisition channel. Purchase decisions increasingly start in AI assistants, so businesses need product feeds and pages that machines can read.
Better personalization. Agents can match products, payment methods, rewards, and delivery terms to user preferences, with strong data governance.
Lower operational costs. In B2B and finance operations, agents reduce manual review, supplier comparison, invoice checking, and payment preparation.
More efficient routing. Agents can choose the payment method by cost, speed, success rate, rewards, or risk, which opens work in orchestration and authorization optimization.
Higher customer lifetime value. A wallet that manages approvals, limits, subscriptions, and purchases across merchants becomes a financial control layer, not only a payment tool.
ilink can provide scalable fintech solutions for your business.

Agentic commerce also introduces new risks.
Unauthorized payment. Businesses must know whether the AI agent had permission to act, what the user approved, and whether the final transaction matched the user's intent.
Fraud. AI agents can be manipulated through fake offers, malicious merchant pages, prompt injection, cloned websites, or misleading product information.
Accountability. If an agent buys the wrong item, exceeds a budget, chooses a fake merchant, or triggers a disputed payment, businesses need clear records showing who authorized what and when.
Data privacy. Agents may need access to preferences, purchase history, payment credentials, delivery details, and business rules. This data must be shared carefully and only when necessary.
False declines and blocked agents. Fraud and bot-protection tools built for human traffic can reject legitimate agent purchases. OpenAI's guidance asks merchants to recognize signed agent requests and work with risk providers to reduce unnecessary declines.
Unclear liability for delegated purchases. In card-based wallets, existing dispute protections still apply. For purchases an agent makes under pre-set limits without approval of each purchase, the rules are still being defined.
The protocols answer these risks in different ways. AP2 focuses on authorization, authenticity, and accountability through signed mandates; Visa and Mastercard rely on agent verification, tokenization, and consumer consent; OpenAI's wallet requires a biometric check for every purchase.
Companies that want to support agentic payments should start with the infrastructure layer. The work differs for those who accept payments and those who issue credentials.
API-first checkout. Agents need structured checkout flows that can be triggered securely through APIs.
Machine-readable product data. Prices, descriptions, availability, shipping, taxes, return policies, and restrictions should be available as feeds (for example, through ACP) or through a UCP manifest.
Agent recognition. CDN, firewall, and bot protection should recognize signed agent requests (Web Bot Auth, Visa Trusted Agent Protocol) instead of blocking them.
Tokenized payment credentials. Accept network tokens and platform tokens; never ask agents for raw card or account data.
Risk rules for agent traffic. Fraud monitoring should tell legitimate agents from abnormal behavior, suspicious merchants, repeated attempts, and manipulated purchase flows.
Audit logs. Every agent action should be recorded, including user intent, approval, cart data, payment token, merchant response, and final transaction status.
Dispute and refund workflows. Agent-led transactions still need clear customer support, chargeback, refund, and cancellation processes.
The API-first pattern already exists in crypto payments. In ilink's White-Label Crypto Processing, checkout runs through a direct API on the client's own website without redirects, and the business keeps control of private keys. Agent-initiated payments need the same two properties: a checkout an API can call, and credentials that never pass through the client application.
Agent-ready tokens. Take part in network programs for agent payments, such as Mastercard Agent Pay and Visa Intelligent Commerce, so that agents receive tokens instead of card numbers.
Strong authentication. Use passkeys or biometrics for approvals and step-up confirmation for high-value, unusual, or regulated transactions.
Delegated limits as a product feature. Let users define limits by amount, merchant, time, category, geography, and risk level, and show them which agents hold access.
Consent and mandate records. Store what the user approved and when, so disputes can be resolved against evidence.
Compliance controls. Depending on the use case, KYC, KYB, AML, sanctions screening, data protection, and transaction monitoring still apply.
Revocation. Users should be able to see and revoke an agent's access at any time.
Most of this sits on top of existing payment gateways supporting cards, bank transfers, and crypto payments, not instead of them.
ilink can build API-first payment gateways, wallets, and crypto processing for agent-initiated payments.

Enterprises should not wait until agentic commerce becomes mainstream. The best approach is to prepare the payment and data foundation now.
Audit the current payment stack. Check whether your checkout, wallet, payment gateway, fraud tools, and reporting systems can support API-led transactions.
Review your product data. Make sure products, services, prices, availability, terms, and restrictions are accurate and structured.
Define agent permissions. Decide what an AI agent can do without approval, what requires confirmation, and what should be blocked.
Add tokenization and secure credential handling. Payment credentials should be protected through tokens, not passed directly through agent systems.
Build approval workflows. Add spending limits, approval rules, role-based permissions, and step-up authentication.
Update fraud monitoring. Traditional fraud systems may not be enough because agent behavior can look different from human browsing.
Start with low-risk pilots. Good first use cases include assisted checkout, subscription management, invoice review, product comparison, and internal procurement.
Measure business impact. Track conversion, approval rates, fraud, disputes, cost per transaction, customer satisfaction, and manual work reduction.
To understand whether agentic commerce is creating business value, companies should track both payment and customer experience metrics.
Agent-led conversion rate;
Checkout completion rate;
Payment authorization rate;
Fraud rate;
Dispute rate;
Average order value;
Cost per transaction;
Customer acquisition cost;
Customer lifetime value;
Manual review reduction;
Time saved in procurement;
Share of transactions initiated or assisted by AI agents.
Agentic commerce will probably grow in phases. In the near term, AI agents will mostly support discovery, comparison, recommendation, cart building, and checkout assistance, with the user approving each payment, as ChatGPT Wallet does today.
Full autonomy will move more slowly because payments require permission, identity, and accountability. The events of 2026 show this order: OpenAI moved its focus from in-chat checkout to discovery in March, and added payments in November only with a biometric check for every purchase.
The long-term opportunity is significant. AI agents may become a common interface for shopping, banking, payments, procurement, subscriptions, insurance, and digital asset management. For fintech companies, the winners will likely be those that combine automation with secure, auditable payment infrastructure.
What is agentic commerce in fintech?
Agentic commerce in fintech is the use of AI agents to support or execute parts of the buying and payment journey. This may include product discovery, comparison, checkout preparation, payment initiation, and transaction approval.
What are agentic payments?
Agentic payments are payments an AI agent initiates on a user's behalf with delegated credentials and permissions. The merchant receives a token instead of the real card or account details, and the user's approval or signed mandate is kept as proof of intent.
Can AI agents make payments?
Yes. As of November 2026, AI agents can pay in ChatGPT Wallet in the US, Mastercard issuers in Europe are enabled for Agent Pay, and protocols such as x402 and MPP let agents pay for services with stablecoins. In most consumer schemes, the user still approves each purchase.
How is agentic commerce different from traditional e-commerce?
Traditional e-commerce is built around human browsing and manual checkout. Agentic commerce adds an AI agent between the user and the merchant, which means payments, authorization, product data, and fraud controls need to work in a more automated and API-driven environment.
What is ChatGPT Wallet?
ChatGPT Wallet is OpenAI's payment feature for agents, launched on November 13, 2026, with American Express, Chase, Mastercard, and Visa. The user approves each purchase with a security check such as Face ID, and the payment network provides a token, so the agent and the merchant never see the real card details.
Why does agentic commerce matter for payment companies?
Payment companies need to support secure agent-led transactions, tokenized credentials, delegated authorization, fraud scoring, merchant verification, and dispute handling.
What is the Agentic Commerce Protocol?
The Agentic Commerce Protocol (ACP) is an open standard developed by OpenAI and Stripe for AI agents and businesses. Since March 2026, OpenAI uses it mainly for product feeds and discovery in ChatGPT, while merchants can keep their own checkout.
What is Google AP2?
Google's Agent Payments Protocol, AP2, is an open protocol designed to support secure agent-led payments. It uses signed mandates to prove user intent and create an audit trail for authorization, authenticity, and accountability.
Who provides agentic payment infrastructure?
The main layers come from card networks (Visa Intelligent Commerce, Mastercard Agent Pay), AI platforms (OpenAI with ChatGPT Wallet and ACP, Google with AP2 and UCP), payment processors (Stripe with Shared Payment Tokens and MPP), and open foundations (x402). Banks, wallets, and PSPs build the control layer on top: limits, consent records, and dispute handling.
What are the main risks of agentic commerce?
The main risks include unauthorized payments, fraud, prompt manipulation, fake merchants, false declines of legitimate agents, unclear liability for delegated purchases, data privacy issues, weak consent, and poor dispute handling.
How can businesses prepare?
Businesses should build API-first checkout, improve product data quality, let verified agents through their bot protection, use tokenized payment credentials, define agent permissions, add approval workflows, update fraud systems, and start with low-risk pilot use cases.
How to build a P2P payment app: product models, launch features, instant payment rails, ledger design, fraud controls, US rules, cost and revenue.
KYB onboarding step by step: business and owner checks, FinCEN's 2026 changes, automated decisions with maker-checker, and how to build it into a product.
ilink can help with custom development and ready-made solutions.
