Former PayPal President: How Will Payment Systems Be Rebuilt in the Age of AI Agents After Apps and Buttons Disappear?

AI agents will buy and sell on our behalf, but today's payment systems still assume there is a person sitting behind a screen tapping "Confirm."

In early 2015, while I was at Meta, the company acquired a small natural language processing firm, Wit.ai, and we began collaborating with the Facebook AI Research team (FAIR). That summer, we tested the mobile assistant M in Messenger. M was an assistant that could message like a friend. You asked it to book a restaurant, send flowers to a loved one, or plan a trip, and it would simply get it done without you touching any app. Users quickly grew fond of it.

At launch, I described M as "artificial intelligence trained and supervised by humans." That was accurate, but we never managed to automate more than 30% of the conversations. At the time, large models hadn't yet emerged; model capabilities were insufficient, resulting in an inability to handle demand. The product was ultimately shut down.

In 2019, we launched Libra, an internet-native digital asset designed to move globally as easily as sending a message. Libra fell victim to political reasons—a story I've told elsewhere. I remain grateful to Mark Zuckerberg for his enthusiastic support of both these ventures.

Both times, I misjudged the timing by a few years. M needed models capable of acting autonomously, and Libra needed governments to recognize digital assets. Now, both have come to pass. The models are ready; stablecoins, which are essentially the prototype of Libra, have already been codified into U.S. and European law.

That two things I bet too early on are now happening simultaneously is worth taking seriously. In my view, combined, they signal the demise of apps and on-screen buttons, paving the way for an entirely new method of moving funds.

The Interfaces We Know Are Disappearing

When I talk about the disappearance of interfaces, I mean that familiar UIs such as apps, websites, and buttons will vanish. You will still naturally look at photos and maps; I am not that extreme. Our way of interacting with AI and computers will become like interacting with people: speak, type, and the task gets done.

For over 40 years, we have been learning the computer's language: programming, menus, forms, dropdowns, hamburger icons, and "Are you sure?" pop-ups. We became so accustomed to it that we forgot we were constantly translating. Now, computers are starting to speak our language.

This is also redefining what an "operating system" is. Today's OS is primarily an app launcher; whoever owns the launcher dictates all downstream rules. When people stop launching apps, the OS becomes the agent you converse with, and it currently has no owner.

Where attention converges will also shift. For the past 20 years, human attention has focused on feeds and apps. Next, it will concentrate on agents: personal agents working for you, enterprise agents working for companies, which will increasingly make decisions on our behalf.

Whoever controls the winning agent will become the most powerful company of this era. I believe there is little room for debate here. Meta's Muse is performing exceptionally well and is only just getting started. Google will likely enter the fray, as will frontier model companies; any company capable of building strong economic network effects among stakeholders will compete. Whether these agents win out, and whether these companies make money, hinges entirely on commerce, since inference, virtual machines, and foundational model training all cost money. Besides, an agent that cannot purchase anything hardly counts as capable.

The Payment System Is Built on Buttons

At PayPal, the product was that button. During my tenure leading PayPal, we spent years optimizing the checkout experience, cutting steps, shaving seconds, and helping more users complete their purchases. Now, I am writing its obituary.

Almost all financial products today are built around a screen: banking apps, checkout pages, card number forms, and verification codes sent to your phone. The entire risk framework for payments—authentication, fraud, disputes, liability—assumes a human is present in many steps. Remove that person, and the entire premise collapses.

When your agent buys from a corporate agent, there is no checkout page and no one to send a verification code. The question becomes: Did you authorize this agent to make such payments? These rails were designed long ago, before anyone needed to ask that question. Card networks and some new protocols are already seeking answers, and everything underneath must catch up.

The interface for digital funds will undergo profound changes because money will no longer flow through a UI.

Stablecoins Are No Longer the Answer

This might sound strange coming from a Libra founder, but I believe stablecoins are no longer a standalone solution.

This framing proved useful. It gave legislators a concrete entity to regulate, and stablecoins were consequently recognized as legitimate digital assets. Europe's MiCA is fully implemented, and the U.S. GENIUS Act has become law, taking effect in January. This battle was fought for years and is essentially won. Stablecoins equal acceptable digital assets.

But stablecoins alone cannot solve most real-world problems. Holding digital dollars is indeed useful. But putting them to use is often difficult. To be useful at scale, stablecoins must integrate with the following:

• Fast settlement networks that deliver value in seconds, anytime, at near-zero cost;
• Accounting systems, or what many today call embedded wallets, giving money a place to reside within the products or agents using it;
• Fiat payment systems, both domestic and cross-border, allowing you to move between all existing systems and effectively siphon liquidity from old rails onto new ones;
• Card networks, enabling balances to be spent wherever cards are accepted, and cards are virtually everywhere.

The Agent-Native Payment Layer

Combine all of this: stablecoin settlement, accounts, fiat rails, and cards, but with no frontend—only a simple API that agents can tap into whenever they need to transfer value, whether buying or selling on behalf of a person or a company. This is the "agent-native payment layer."

The API is simple, but the backend is not. You need licenses, banking partners, global issuance and acquiring, real-time connections to local payment systems in over 100 countries, liquidity, compliance, and everything operating simultaneously. This is the hardest, most grueling work in payments, which is why so few companies succeed.

You tell your agent to fix the fence before the weekend. It finds a contractor—or more likely, the contractor's agent, negotiates the price, pays a deposit, and waits for photos of the completed job before paying the balance. You never touched an app or went through a payment workflow. You just stated what you wanted in a single sentence.

Consider a company whose agent manages a treasury and a set of rules: paying 400 contractors across 30 countries every Friday, scaling ad spend when data looks good, paying cloud invoices, and issuing refunds to customers who received damaged orders. Every transaction hits a different endpoint: a bank account in this country, a mobile wallet in that country, a card, a stablecoin address. The agent shouldn't care about any of that. It should just say "pay this person this amount," and the funds arrive within seconds, anywhere, around the clock.

Agents have their own inboxes, some even have their own phone numbers, but the current AI economy lacks banks, and agents lack accounts. To give agents accounts, the underlying money must:

• Be digital;
• Settle in real time, around the clock, regardless of banking hours;
• Be able to reach every endpoint on Earth;
• Be cheap enough that an agent pays mere cents per API call, ensuring fees do not exceed the payment itself;
• Allow humans to safely grant agents actual spending authority with set limits that can be revoked instantly;
• Have a clear answer for who bears the loss if an agent is tricked into making the wrong payment.

Who Will Build the Agent-Native Payment Layer?

Payment infrastructure will spawn a batch of giants, several of which will reach market caps in the hundreds of billions. By that momentum, those who win the agent race will see that figure as mere change.

The gap comes down to how each side makes money. This payment layer earns 1 to 300 basis points on volume, meaning 0.01% to 3%. An agent that decides what buyers purchase and from whom is worth infinitely more to sellers than mere capital transporters. The companies running these agents will be compensated similarly to today's marketplace and ad platforms.

This does not mean payments are unimportant within agent platforms. On the contrary, their payments teams must tackle one of the company's hardest challenges: piecing together licenses, banks, card networks, local payments, and compliance into a secure, user-friendly experience that works in every country and serves both buyers and sellers. The underlying funding layer will become as reliable and critical as the protocols that run the internet.

For those who execute on the agent-native payment layer, this represents a massive opportunity.

This Time, It May Not Be Too Early

The first wave of agent shopping started slowly, and I can already hear voices saying I am once again too early. I believe money is a major factor. Agents have been piggybacking on payment flows built for humans, with every purchase ultimately ending in a card form and a confirmation button.

This time, the models are good enough, and the laws are written. What is missing is money that agents can use without a screen. Whoever builds it properly will construct the foundation everything else relies on. If I am ahead of the curve again, so be it.

Join the official Coincamps community:

X: https://x.com/coincamps

Telegram: https://t.me/coin_camps