Original Title: Agentic Commerce Won’t Kill Cards, But It Will Open a Gap

Original Author: Noah Levine, a16z Investment Partner

Compiled by: Saoirse, Foresight News

A few weeks ago, an article published by Citrini Research claimed that stablecoins would bypass Visa and Mastercard, directly causing a sharp decline in the stock prices of card networks. The crypto community cheered.

This logic sounds clear: AI agents will optimize every transaction, transaction fees are a type of ‘tax,’ and stablecoins can bypass it.

I spend all day in the crypto space and also wish this argument were true, but most of it is wrong.

It’s not because stablecoins are unimportant, but because the real opportunity isn’t about replacing bank cards at all—it’s about serving merchants who have difficulty accessing traditional card payments.

Bank Cards Will Capture the Vast Majority of the Market

Citrini’s argument is built on an assumption: AI agents, free from human habits, will actively optimize away card network fees.

But bank cards are more than just transfer tools. They provide unsecured credit, pre-authorize uncertain transactions, and offer fraud protection through chargeback rights.

Stablecoins can transfer funds, but they can’t do the rest.

Imagine your agent books a hotel for you, but it turns out to be nothing like the pictures.

With a bank card, you can dispute the charge and get your money back.

With stablecoins, once the money is sent, it’s gone for good.

82% of Americans hold reward credit cards (referring to credit card perks like cashback, points, airline miles, hotel points, etc.), and there are over 18 billion cards in circulation globally.

For the vast majority of transactions, consumers won’t voluntarily give up purchase protection and rewards to choose a payment method that offers no benefits and is irreversible.

Fraud detection is another huge advantage of card networks: they can run models on billions of transactions in real-time.

Stablecoins currently lack a comparable network-level anti-fraud layer.

Microtransactions are often cited as a weakness of bank cards, but card networks have long adapted to such mismatched transactions.

Visa has processed over 2 billion transit ticketing transactions by aggregating multiple swipes into daily settlements.

The card industry has never abandoned any type of transaction; it always invents new products to cover them.

Another challenge is: “Agents can’t hold cards.”

But agents are essentially just new devices.

Your phone, watch, and computer all hold independent tokens pointing to the same card, just like Apple Pay.

Your phone never went through KYC; it just holds your token. Agents will be the same.

Visa has issued over 16 billion tokens, and agents will use these tokens too.

Visa’s Smart Commerce framework is in pilot, and Mastercard’s Agent Pay is already live for all U.S. cardholders.

The agent commerce protocol built by Stripe and OpenAI is already integrated with Etsy, and over a million Shopify merchants are about to go live.

The conclusion is clear:

For existing merchants and consumers, bank cards are almost certain to dominate agent commerce.

The opportunity for stablecoins lies elsewhere—with merchants that haven’t even emerged yet.

Those Merchants That Haven’t Emerged Yet

Every platform shift gives rise to a wave of merchants that existing payment systems cannot serve.

When eBay emerged, individual sellers couldn’t open merchant accounts; PayPal served them;

Shopify grew from 42,000 merchants to 5.5 million over 13 years;

When Stripe was founded, many of its clients hadn’t even been born yet.

The pattern has always been consistent: the winners serve merchants that existing giants cannot underwrite.

The AI wave will generate such merchants faster than any previous platform shift.

Just last year, 36 million new developers joined GitHub.

In YC’s Winter 2025 batch, a quarter of the companies had codebases with over 95% AI-generated content.

On the popular AI coding platform Bolt.new, 67% of its 5 million users aren’t developers at all.

People who couldn’t write production-level code two years ago are now releasing software.

They are both buyers of developer services and sellers at the same time.

Imagine this:

An ordinary developer uses AI tools to spend 4 hours building a tool that displays financial data for public companies. No website, no terms of service, no legal entity.

Another developer’s agent calls it 40,000 times a week, at 0.1 cent per call, generating $40 in revenue. No one ever clicks a checkout page.

I see developers building tools like this every week.

Their first question is always: How do I get paid?

For most, the answer is: You can’t, right now.

Existing payment institutions struggle to onboard such merchants.

It’s not a technical issue; it’s that once a payment institution onboards a merchant, it assumes their risk.

If the merchant commits fraud or generates a large number of chargebacks, the payment institution takes the blame.

A tool with no website, no entity, and no track record almost never passes risk control reviews.

The system is working as designed—it just wasn’t designed for this scenario.

Payment institutions can certainly adapt; they’ve done it before.

But it took PayPal 16 years from launch to the industry’s first underwriting guidelines for payment service providers.

And these new merchants need to get paid now.

For them, accepting stablecoins is like a street vendor only taking cash.

It’s not that cash is better; it’s that such merchants have historically struggled to get approved for card acceptance.

In this gap, stablecoins are currently the only viable solution.

Despite rough wallet experiences and still-forming compliance frameworks, protocols like x402 can embed stablecoin payments directly into HTTP requests:

No merchant account needed, no processor, no onboarding, no chargeback liability.

These merchants aren’t choosing between stablecoins and bank cards.

They’re choosing between stablecoins and not getting paid at all.