How to Add Payments to Your AI Agent Without Becoming a Payments Company
Key takeaways
- An agent that charges the user and then buys from the merchant makes its company the merchant of record for everything the agent buys.
- When a platform collects user money and pays merchants with it, the platform is in the flow of funds, with licensing, tax, refund, and working capital consequences.
- Three build paths put a platform in that position: becoming merchant of record, running a card issuing program, and building per-merchant integrations with their own commercial terms.
- In a pass-through model, the user's own tokenized card pays the merchant directly under terms the user approved, so the platform never holds funds.
- When the user pays with their own card, the user keeps the normal protections of their card, and the platform is not the party in a dispute.
- The choice is cheapest before the first ledger table exists.
You add payments to an AI agent without becoming a payments company by keeping your platform out of the flow of funds. In a pass-through model, the user's own card is tokenized and charged by the merchant under terms the user approved, so your platform never collects, holds, or forwards money. Your company owns the product experience, and the merchant stays the seller.
This article is engineering and product guidance, not legal advice. Whether a specific model triggers licensing or tax obligations depends on your structure and jurisdictions, and your counsel should decide.
Why can't I just add Apple Pay to my agent?
You cannot just add Apple Pay because Apple Pay pays the seller in the transaction, and inside your agent that seller is your company. The user confirms, the money lands in your account, and your company then has to buy the item from the real merchant. That step makes you the merchant of record.
What builders usually want from Apple Pay is the experience: a native sheet, a clear amount, and one biometric confirmation. That experience is separate from the payment rail under it. An approval screen can look and feel the same while the credential is the user's own card and the seller on the transaction stays the merchant.
What does "becoming a payments company" mean?
Becoming a payments company means your platform takes user money and pays it out to third parties, which brings the obligations of a business that moves money. It usually starts with a shortcut, such as a stored balance or a charge to the user followed by a purchase on the user's behalf.
- Licensing exposure: Money transmission rules are written to cover businesses that accept funds and pay them to third parties, which can require analysis state by state and country by country.
- Tax and reporting: When your company sells to the user, sales tax, invoicing, and information reporting become your obligations instead of the merchant's.
- Refunds and chargebacks: When your company name is on the card statement, the dispute comes to you, including for merchant failures you did not cause.
- Working capital: Your company pays between the charge to the user and the payment to the merchant, and that float grows with volume.
- Compliance staff: Customer identification, sanctions screening, transaction monitoring, and audits need people as well as code.
Why does a merchant-of-record model put you in the flow of funds?
A merchant-of-record model puts you in the flow of funds because your company becomes the seller: the user buys from you, and you buy from the merchant. That model suits a business that resells its own inventory. It is a heavy model for an agent that buys many kinds of things from many stores.
A traditional travel agency is a familiar example. When a customer pays the agency and the agency buys the ticket, the customer's relationship is with the agency, and so are many of the tax, refund, and consumer protection obligations. An AI agent that charges users and then buys inherits the same position, in every category at once.
Agent purchases have no fixed category. One consumer agent can order dinner, buy a phone charger, and book a hotel on the same day. As merchant of record, the agent company becomes a reseller of restaurant food, electronics, and travel, each with its own tax treatment and refund norms.
Grocery delivery services show the operational side of this model. The customer pays the service, the service buys from the store, and the service operates support for substitutions, refunds, and delivery problems. That works when delivery is the whole business. It is a large commitment to take on as a side effect of adding payments to a chat product.
What changes when an AI agent does the errand?
When an AI agent does the errand, the agent no longer has to be the merchant: the merchant can stay the merchant of record while the agent does the work. Earlier intermediaries had to sit in the middle of every transaction because the only way to buy on a customer's behalf was to take the customer's money.
In a pass-through model, the user pays the merchant directly with their own card, under terms they approved, and the agent company never enters the flow of funds. The user's statement shows the store they bought from, their card protections apply directly, and their order history and recourse sit with the real merchant.
What are the do-it-yourself paths, and what do they cost?
The three do-it-yourself paths are becoming merchant of record, running your own card issuing program, and building per-merchant integrations. Each one works, and each one turns part of an engineering roadmap into a payments roadmap.
Path one: become merchant of record
You charge the user and then buy from the merchant. This is the fastest path to a demo and the most expensive path to operate.
- What you build: A ledger, a payments account, refund handling, dispute response, tax calculation, and reconciliation between what you charged and what you paid.
- What you inherit: A position in the flow of funds, chargeback exposure for merchant failures, sales tax obligations, and float.
- How long it takes: Months to start, and the work continues, because each new purchase category adds tax and refund cases.
Refunds show the cost clearly. When a user wants money back, you usually refund the user first and then recover the money from the merchant later. Settlement timing adds a second gap, because you pay the merchant at purchase and receive the user's funds days later. Your company finances both ends of each transaction.
Path two: run your own card issuing program
You issue virtual cards and let the agent spend them. The technical work is lighter than merchant of record, but someone must fund the cards, and that party is your company.
- What you build: A funding mechanism, a process to collect from users, and controls that stop an agent from overspending a card.
- What you inherit: Card program compliance, prefunding requirements, and the risk of paying before you collect.
- What you still do not have: A way to complete a merchant checkout. A card number does not sign in, pass a challenge, or submit a form.
A new single-use card for every task can also resemble card testing to merchant risk systems: many short-lived numbers, small amounts, and correlated timing. Valid cards can be declined because of the pattern.
Path three: build per-merchant integrations
You integrate with merchants one at a time. This gives the best experience at each merchant and the slowest growth in coverage, because each merchant is a separate relationship, a separate API or partnership, and separate maintenance.
Consumer demand often concentrates on a small number of large merchants. Those merchants are also the least likely to build a custom integration for a small platform, so negotiated coverage often reaches smaller merchants first.
Each do-it-yourself path adds payments operations to a team that set out to build an agent.
Where does the reseller model work well?
The reseller model works well in travel, where aggregators and global distribution systems have packaged airline and hotel inventory for decades. That layer lets an online travel agency sell against one counterparty, so an AI travel agent can plausibly act as merchant of record for flights. General commerce has no equivalent layer, so an agent that wants to be the seller outside travel has to onboard merchants one by one, which makes it a marketplace.
A promise to buy anything for the user also has no natural boundary. The same process that buys a phone charger is responsible for a piano, and the float and refund exposure grow with every purchase.
How does a pass-through payment model work?
A pass-through payment model works by spending the user's own tokenized card directly at the merchant, under terms the user approved, so the platform is never a party to the movement of funds.
- 1
The user links an existing card: A PCI DSS compliant vault tokenizes it. The agent never sees the card number, and no stored balance exists.
- 2
The agent submits an intent: The user's request becomes a structured purchase record with merchant, items, prices, and a maximum total.
- 3
The user approves it in your product: An approval screen renders in your own interface and the user confirms with a passkey or biometric, as with Apple Pay.
- 4
The purchase executes at the merchant: The system completes the checkout as an identified agent and checks each step against the approved terms.
- 5
The charge belongs to the user and the merchant: The user's bank sees a purchase the user authorized. Your platform never collected, held, or forwarded funds.
Because the charge is between the user and the merchant, refunds and disputes follow the normal card network process, and the user keeps the protections of their own card.
How do the four options compare?
The four options differ most in whether your platform holds funds, whose name is on the statement, and who answers a dispute.
| Merchant of record | Own issuing program | Per-merchant integrations | Pass-through model | |
|---|---|---|---|---|
| In the flow of funds | Yes | Yes, you prefund | Depends on each agreement | No |
| Name on the statement | Your company | Your program's card | Varies | The merchant |
| Refund and dispute exposure | Yours | Largely yours | Shared | Normal card network process between user and merchant |
| Working capital required | Significant float | Prefunding | Low | None |
| Compliance program | Yes | Yes | Limited | Limited, because the platform holds no funds |
| Completes merchant checkout | You build it | No | At each integrated merchant | Yes, where the execution layer supports the merchant |
| Merchant coverage | What you build | What you build | Only signed merchants | Depends on the execution layer |
What do you still own in a pass-through model?
In a pass-through model you still own the conversation, the interface, the customer relationship, and the economics. The parts that users judge a product on stay with the builder.
- The conversation: How your agent understands a request, asks clarifying questions, and makes recommendations.
- The interface: Approvals render in your own product, with your own design.
- The relationship: The user is your user, in your branding and your support flow.
- The economics: You decide whether purchasing is a free feature, part of a subscription, or a fee, without a payments profit and loss statement.
How do you know which model you are in?
You know which model you are in by asking whose bank account a refund leaves. If the refund leaves your account, you are in the flow of funds and the obligations in this article apply. If it leaves the merchant's account, you are a pass-through.
A second test is the card statement. Your company name means your company sold something. The merchant's name means the user bought something and your agent helped.
Make this decision before you write the first ledger table. If your company does not intend to be a payments company, keep the platform out of the flow of funds from the start.
Frequently asked questions
How is an AI agent different from a delivery or shopping service?
Many delivery and shopping services are the merchant of record: the customer pays the service, the service buys from the store, and the service holds the money and the liability. An AI agent on a pass-through model does the same errand without holding funds, because the user pays the merchant directly with their own card.
Can I just use Apple Pay in my AI agent?
Apple Pay inside your agent pays your company, which makes your company the merchant of record for what the agent buys, with the tax, refund, and float obligations that follow. A pass-through model can give the same native, one-confirmation experience while the user's own card pays the merchant.
How can an AI agent make purchases without the company handling money?
An AI agent can make purchases without the company handling money by using a pass-through model. The user's own card is tokenized and charged by the merchant under terms the user approved, so the charge belongs to the user and the merchant, and the platform needs no ledger or settlement process.
Does an agent platform need a money transmitter license?
That depends on the platform's structure and on legal analysis in each jurisdiction. The risk comes from accepting user funds and paying them to third parties. A platform that never holds user funds, because the user's own card pays the merchant directly, reduces that exposure.
What is the problem with becoming merchant of record for agent purchases?
Merchant of record makes your company the seller, which brings sales tax obligations, refund and chargeback exposure for merchant failures, and working capital to pay for purchases first. For an agent that buys in many categories, the company becomes a reseller of food, electronics, and travel at the same time.
Who handles refunds and disputes in a pass-through model?
In a pass-through model the charge is between the user and the merchant, so refunds and disputes follow the normal card network process. The user keeps the protections of their own card, and the agent platform is not the disputed party.
Why not build integrations with the merchants my users care about?
Per-merchant integrations give a good experience at each merchant but grow coverage slowly, because each merchant needs its own agreement, integration, and maintenance. The largest merchants are also the least likely to build a custom integration for a small platform.
Related articles