For years, the most AI could do in digital commerce was help people choose. It surfaced alternatives, summarized reviews and helped figure out which product was worth the money. That was useful. It was not a structural shift.
The next phase is something different. Agents are beginning to act, not just advise. And that changes every relationship in the transaction.
Instead of pointing a consumer toward a product, an AI agent can now search for the item, verify eligibility for a discount, generate a payment credential and complete the purchase. The consumer never touches the checkout page. McKinsey project that model could represent between $3 trillion and $5 trillion in global commerce by 2030.
Also read: Elon Musk delivers 6-word warning on future of employment
Payment networks face a question they were not built to answer
The payment system was built for one thing: a person pressing a button. Card networks and banks have always asked two questions: is this credential valid, and does this transaction look legitimate? AI agents add a third. Is the software making the purchase actually allowed to spend this person’s money?
“The biggest change is that the customer’s intent, the purchasing decision and the payment can increasingly happen through an agent rather than a person navigating a website,” Kenneth Shek, CEO of Moca Network and Head of Projects at Animoca Brands, told The Street in an interview.
That is exactly the problem Animoca Brands set out to solve. Working with fintech companies Reap and Pivota, the company is building infrastructure that lets an agent verify a user’s eligibility, locate a product and generate a payment credential locked to a specific merchant, amount and time window.
Animoca’s AIR identity layer and Minds AI agent connect to Reap’s payments stack and Pivota’s live pricing data. A controlled merchant showcase comes first, with a live pilot using real transactions to follow.
Payment providers will need to think in finer increments. A blanket approval to spend $200 is not permission to spend it anywhere, at any time. Visa and Mastercard are already building frameworks for exactly that distinction. But most of the infrastructure for machine-initiated transactions is still being designed.
Owning the shopping experience is not the same as owning the pipes beneath it.
The payment credential may need to be rebuilt for machines
The standard payment card was designed for a person. It was never meant for software, and the difference matters when the stakes involve someone’s real money.
“Giving an agent permission to shop should not mean handing it an unrestricted card,” Shek said. “The authorization should be specific: which user it represents, what it can buy, which merchant it can pay, the maximum amount and when that permission expires.”
In that model, a purchase-specific credential is generated for each transaction, scoped to a single merchant, amount and time window. If anything changes, the transaction fails. The agent cannot reinterpret the user’s instructions to make the numbers work.
Nate Herk, CEO of AI Automation Society, learned that lesson firsthand. An agent his team built once misread a task. Instead of drafting an email, it sent one, pushing a discount code out to 150,000 people.
“Payments work the same way,” Herk told The Street in an interview. “Make every purchase need a human’s approval, or cap spending per week or month. Put the rule in the card, not in the prompt.”
The limits on what an agent can do should not depend on the agent remembering them correctly. They need to live at the infrastructure level, where the money actually moves.
Whoever controls the agent controls the purchase decision
The money question is just the start. If agents become how people shop, whoever runs the agent gets to decide what they see. Which products appear, which merchants get recommended, which payment method gets used. That is an enormous amount of commercial influence concentrated in a single interface.
“Whoever owns the assistant people shop through has the most leverage,” Herk said. “That company decides which products you see and which checkout gets used, which is Google’s search leverage carried all the way to the purchase.”
More AI:
- Nvidia just made a move Wall Street wasn’t ready for
- Microsoft just took sides in AI policy fight
- OpenAI just disclosed something genuinely alarming
Merchants already know what it means to depend on a platform for visibility. Google’s dominance in search showed how quickly discovery becomes a toll road. Getting recommended by an AI agent may end up working the same way.
And consumers will face their own version of the problem. An agent with commercial relationships may not always surface the product that is actually best for the buyer.
The financial rails underneath could still be contested
Owning the shopping experience is not the same as owning the pipes beneath it. Banks, card networks and payment processors would still do what they have always done: authorize, settle, handle fraud and resolve disputes. But more of the transactions crossing their rails would come from machines they never vetted.
Sachi Kamiya, Director of Venture and Growth at Sentient Foundation, sees that gap as a real opening for new competition.
“AI platforms could control the consumer interface, while payment networks, banks and wallet providers compete to process the transactions behind it,” Kamiya told The Street in an interview.
Blockchain is one of the technologies competing for that infrastructure role. AI agents can use funded wallets to transact in stablecoins, moving value without relying on every intermediary a conventional payment touches. That does not mean blockchain becomes the default. But it gives AI companies an alternative as they build machine-to-machine commerce.
“My hope is that this competition reduces traditional financial institutions’ control over payments,” Kamiya said. “Blockchain could play an important role by enabling faster settlement and potentially lower transaction costs, although fees and safeguards vary across systems.”
Identity becomes the foundation of the whole system
Strip away the technology and the same question keeps coming up: who does this agent actually work for? The merchant needs to know it represents a real buyer.
The payment provider needs to know it has authority over the account. And the buyer needs to know the agent is staying inside the limits they set, not inventing new ones on the fly.
One answer is to keep those functions in separate layers rather than bundling them. Identity, intent, product discovery and payment each handled by a different part of the system. That separation could allow users to move between agents without losing financial permissions, membership status or purchasing history.
That portability matters more than it looks right now. A consumer who gives an agent payment authority and earns benefits through it will think twice before switching. The company holding those credentials has real leverage.
In a market where AI platforms are competing for the same consumer, the one that earns trust first may be the hardest to walk away from. When the agent is spending your money, the question of trust is not abstract anymore.
