# The Death of the Interface

Source: https://x.com/davidmarcus/status/2109690025336242608

## Summary

The author argues that apps, websites and buttons are dying, and that money will move through agents via "headless money," meaning stablecoin settlement, accounts, fiat connectivity and cards exposed through a simple API with no frontend. The article ties this to M, a Messenger assistant that Meta tested in 2015 and later wound down after automating no more than 30% of interactions, and to Libra, which was killed for political reasons. The author says models and stablecoin laws, including MiCA in Europe and the GENIUS Act in the U.S., now exist, so the timing is right. The author also says the author's company, Lightspark, is building headless money.

## Article

What happens to money when apps, websites and buttons go away

I generally tend to be early with things. On this one, I managed to be early on both halves: the interface with M and the money with Libra. This time around, I feel right on time.

In early 2015, at Meta, we acquired a small natural language startup called Wit.ai, and we started working with the Facebook AI research team (FAIR). That summer, we started testing M inside Messenger. M was an assistant you could text like a friend. You'd ask it to book a table, get flowers to your loved one, or sort out a trip, and it would just get it done without you ever touching an app. People loved it instantly.

When we announced it, I described M as “artificial intelligence, trained and supervised by people,” which was true, but we never managed to automate more than 30% of interactions. This was the pre-LLM era. The models weren't there yet. As a result, we couldn't scale to meet the demand, and we eventually had to wind down the product.

In 2019, we announced Libra, money native to the internet that could move around the world as easily as a message. Libra was killed for political reasons, a story I've told elsewhere. I'm still grateful Mark (Zuckerberg) enthusiastically backed both swings.

Both times, I got the timing wrong by years. M needed models that could do the work on their own, and Libra needed governments to accept digital money as real money. Both of these have now happened. The models are good enough, and stablecoins, which is basically what Libra was, are now written into law in the U.S. and in Europe.

When two things you were early on show up at the same time, you pay attention. I think together, they mean the death of the interface and a whole new way for money to move that will come with it.

The interface is dying

What I mean by the death of the interface is the death of UI as we know it. Apps, websites, buttons, everything goes away (you'll still look at photos and maps, I'm not that extreme…). We're going to deal with intelligence and computers the same way we deal with other people, by talking and typing, and things will just get done.

For over 40 years, we've been learning to speak the computer's language: coding, menus, forms, dropdowns, the hamburger icon, the "are you sure?" pop-up. We got so used to it that we forgot we were the ones doing the translating. Now the computers speak our language.

That changes what we call an operating system. Today, an OS is mostly a launcher for apps, and whoever owns the launcher sets the rules for everything downstream. When people stop launching apps, the operating system becomes the agent they talk to, and nobody owns that yet.

It also changes where attention aggregates. For the better part of 20 years, human attention pooled in feeds and apps. Now it's going to pool in agents, personal ones working for us and business ones working for companies, and they'll make more and more decisions on our behalf.

The companies that control the winning agents will be the most powerful companies of this era. I don't think that's a controversial take anymore, and you can already see who's going after it. Meta's Muse is an exceptionally well-built product, and it's still incredibly early. In the startup world, there's Instinct. Google will likely play here as well, and so will the frontier model companies, and any company with large economic network effects between their stakeholders.

Commerce will be a huge part of how these agents win and how these companies monetize and afford the cost of inference, virtual machines, and training the underlying models. Also, an agent that can't buy anything is not a very capable agent.

Money was built for the button

At PayPal, the product was the button. When I was running it, we spent years making that button experience better, shaving off steps and seconds so more people made it through checkout, and now here I am writing its obituary.

Pretty much every financial product we use today is built around a screen: the banking app, the checkout page, the card form, the code texted to your phone. The whole risk stack of payments (authentication, fraud, disputes, and liability) assumes a human in the loop at many steps.

Take the human out, and that assumption completely breaks. When your agent buys something from a business’s agent, there is no checkout page and nobody to text a code to. The question becomes whether you authorize this agent to make this kind of payment. Today's rails were designed long before anyone needed to ask that question. Card networks and a few new protocols are already working on answers, and everything underneath will have to catch up.

The interfaces to money are about to change profoundly. Money is simply going to stop moving through UI.

Stablecoins aren't a thing anymore

This might sound strange coming from the guy who started Libra, but I don't think stablecoins are a thing anymore.

The term is useful. It gave lawmakers something they could get their arms around and regulate, and that's how stablecoins got recognized as legitimate digital money. Europe's MiCA went into full force this year, and in the U.S., the GENIUS Act became law last year and takes effect in January. That fight took years, and it's largely won. Stablecoins = acceptable digital money.

But on its own, a stablecoin isn't enough to solve most real-world problems. Holding digital dollars is a genuinely useful thing (ask anyone in Buenos Aires or Lagos), but doing anything with them is where it often gets hard. To be useful at scale, stablecoins need to be combined with:

Fast settlement networks, so value moves in seconds, any time of day, for close to nothing.

Account systems, what most people now call embedded wallets, so money has a place to live inside whatever product or agent is using it.

Fiat payment systems, domestic and global, so you can move between every system that exists today and basically suck all of the liquidity from the old rails onto the new ones.

Card networks, so a balance can actually be spent everywhere cards are accepted, which is nearly everywhere.

Headless money

Headless money is all of that (stablecoin settlement, accounts, fiat connectivity, and cards) with no frontend, just a simple API surface that agents can readily use whenever they need to move value, whether they're buying or selling goods and services on behalf of people and companies.

The API is simple, but what's behind it isn't. You need licenses, bank partners, global card issuing and acquiring, real-time connections into local payment systems in 100+ countries, liquidity, and compliance, all working at once. That's the "eat glass" part of payments, and it's why very few companies will be able to pull it off.

You tell your agent the fence needs fixing before the weekend. It finds a contractor, or more likely the contractor's agent, agrees on a price, pays a deposit, and releases the rest when a photo of the finished fence comes in. You never interacted with an app or a payment flow. You just asked what you needed with one sentence.

Or take a business whose agent controls a pool of money and a set of rules: pay 400 contractors in 30 countries every Friday, top up ad-spend when the numbers justify it, pay the cloud bill, refund the customer whose order showed up broken. Every one of those lands on a different kind of endpoint: a bank account in one country, a mobile wallet in another, a card, a stablecoin address. The agent shouldn't have to care. It should be able to say, "pay this person this much," and have it happen in seconds, anywhere, 24/7.

Agents have their own inbox, some have their own phone number, but right now the AI economy has no bank, and agents have no account. For them to get one, the money underneath has to:

be digital

settle in real time, all the time, not on banking hours

reach every kind of endpoint on earth

be cheap enough that an agent can pay a few cents for an API call without the fee costing more than the payment itself

let a person safely give an agent real spending authority, with limits, and revoke it instantly

have a clear answer for who eats the loss when someone tricks an agent into paying the wrong party

Who builds the money layer

The money stack is going to produce some very large companies. A few of them will be worth hundreds of billions of dollars, which, at the rate things are going, will look like pocket change to whoever wins the agent race. And that’s ok.

That gap comes down to how each side makes money. A payment stack earns somewhere between 1 and 300 basis points on what moves through it. An agent that decides what a buyer buys and from whom is worth far more to the seller than whoever moves the money, and the companies running these agents will get paid accordingly, the way marketplaces and ad platforms do today.

None of that makes payments less important inside the platforms building these agents. Their payments teams will have one of the hardest jobs in the building, stitching all of these pieces together into a safe experience that just works, in every country, for every kind of buyer and seller. And the money layer underneath will become as dependable and important as the protocols that run the internet.

That's a huge opening for whoever builds headless money well. Full disclosure, that's what we're working on at Lightspark, so I'm biased. But I'd still be making the same argument if we weren't.

Not early this time

The first wave of agent shopping has had a slow start, and I can already hear people saying I'm early a third time. I think the money is a big part of why. Agents have been borrowing payment flows built for humans, with a card form and a confirm button at the end of every purchase.

This time the models are good enough and the law is written. What's missing is money agents can use without a screen, and whoever builds it well will have built something every agent depends on. If I'm early a third time, I can live with that!