a16z: Why do AI agents require stablecoins for B2B payments?

By SamBroner, a16z Crypto

Translated by AididiaoJP, Foresight News

As a tourist walking through a bazaar, you might witness scenes like this: people bustling around, staring at goods, comparing items, sampling samples, bargaining with vendors, and paying. It looks like a series of one-off transactions, with each interaction being a small negotiation, trust mediated by cash in hand or value exchanged via bank cards.

But this isn’t how most business is done in the bazaar. Look closer: most people are locals, walking purposefully toward their favorite vendors. Restaurant owners visit their friends: butchers, fishmongers, and farmers. Tailors go to see mechanics, weavers, and artisans. Both sides extend credit and pay later.

When we talk about how smart agents will make payments, we default to thinking like tourists.

But smart agents will act like locals. The features that make smart agents different from humans—unlimited copying, flexible resource allocation, zero startup costs—mean that a small number of smart agents can dominate niche markets. Even as building smart agents becomes easier, relationships, partnerships, and trust still help create winning experiences. Dominant smart agents won’t need tourists’ payment rails. They need vendor relationships, working capital, and credit.

What might this look like? As smart agents integrate into platforms resembling businesses, payment methods must shift from retail payment rails to pre-negotiated B2B terms and credit, which current rails can’t fully support. This is an opportunity window for next-generation payment rails, like stablecoins, if entrepreneurs can develop excellent solutions for future payment scenarios—such as smart agents, streaming payments, and high-volume, low-value global transactions.

This article explores this idea in three parts: how smart agents differ from humans and how these differences shape successful payment strategies; why current methods fall short; and what needs to be built for the next-generation payment rails to succeed.

How Smart Agents Differ from Humans

To understand smart agents and payments, we must consider two questions: Will smart agents behave like humans or like businesses? Will they play long-term or short-term games?

Smart agents will behave more like businesses, establishing long-term relationships with vendors and partners. They will be instances of larger enterprise structures—like a perfect tour guide in a well-connected travel agency, or a franchisee adjusting operations to local tastes without renegotiating supply chains.

Why will smart agents act like businesses?

First, the best experiences are carefully designed. I don’t want an agent that’s still browsing, comparing prices, and bargaining at checkout. I want an agent that has already done those things—knows which vendors are reliable, has pre-negotiated prices, and can settle immediately. This is a business relationship, not a tourist transaction.

In fact, human agents already exist: travel agencies, literary agents, talent agents, watch distributors, real estate brokers, and more. These agents build key multi-round relationships—working with publishers, studios, watch distributors, or mortgage lenders—each transaction tailored on that basis.

Second, while smart agents are infinitely replicable, scaled enterprises (and their advantages) are not. The best smart agents leverage the cost and benefits of scale: cheaper compute, better vendor pricing, deeper integrations, and more predictable components. Scale brings scale: a travel agency booking a million flights annually will get better terms from airlines than one booking only ten.

We’ve seen this before. Only companies like ChatGPT have direct negotiation channels with platforms like Shopify, Amazon, Expedia. Small startups are constrained to using automated browsers or reverse-engineered APIs, paying retail-like fees.

That’s why smart agents will consolidate, or at least most will be built on larger platforms. Building a smart agent is easy, but economics favor only a few winners per vertical—each with deep vendor relationships and profit margins to reinvest in better experiences. Vertical-specific agents with strong vendor ties can work alongside user agents to deliver a win-win experience.

Two Payment Relationships

If smart agents act like businesses, then two payment relationships must be designed: user → agent, and agent/platform/“tour guide” → vendor.

Users pay the agent—via subscription, task-based fees, credit lines, or granting access to their accounts. The agent then pays vendors through pre-negotiated B2B terms, bulk pricing, net 30 invoices, or sub-agents. Currently, agents occasionally pay vendors via retail rails, but this is only a small part of overall spend.

This is essentially how credit cards work today: issuers have retail relationships with consumers, bearing risk, creating customized rewards, and offering credit. Acquirers have business relationships with merchants, including negotiated terms, bulk transfers, and complex working capital dialogues.

Agents and Credit Cards: A McKinsey-Style Perfect Match

As many have noted, credit cards are a surprisingly reasonable payment product for agent use cases. They are widely accepted; payments between $20 and $1,000 are considered reasonable; and they come with arbitration, cancellations, and digital features.

Credit cards also have monthly billing—a key opportunity for consumers to understand what they’ve paid. As smart agents replace kids playing with iPads and causing unexpected expenses, this concept will evolve.

But there are two issues: first, credit cards are technically ill-suited for agents; second, their fee models push the industry into a classic innovator’s dilemma.

Credit Card Technology Is Difficult to Upgrade

Almost all credit card tech assumes human involvement: an approval step, a user interface layer, and traditional payment types (single payment, subscription). Products like Stripe Link, Visa 3D Secure, and dozens of virtual card solutions—allowing you to save cards on websites or register cards for recurring monthly payments—have finally matured after over 15 years.

The adoption of agent payments is happening too fast for thousands of PSPs, POS systems, merchants, and endpoints to upgrade their interfaces, programmability, and fraud detection systems slowly.

Credit cards are inadequate for high- and low-cost payments

Imagine an agent streaming funds to a compute provider or micro-paying for API access. Neither can run on the credit card rails. Visa doesn’t support payments below a cent; its fee model expects a flat 30 cents per transaction. While Visa could develop streaming or micro-payment tech, convincing stakeholders to accept lower revenue would be very difficult.

More problematic, credit cards are caught in an innovator’s dilemma. Despite similar user relationships and payment needs, agent payments often fall outside the $20–$1,000 range. Worse, many initial scenarios involve API fees that are hard to refund or resell (fraud).

Even beyond credit cards, traditional rails will still have a place in the future.

Existing Payment Methods Still Have a Role

As smart agents become integrated into platform-like structures, most high-volume spending will shift to pre-negotiated B2B terms: invoices, net 30 payments, discounts, and credit lines. In that world, “payment rails” can be anything—often just a simple asynchronous settlement on traditional rails. Costs are spread across larger transactions, and working capital is negotiated between companies.

But agents won’t only exist in that world. They are already emerging in areas where traditional payment methods perform poorly: initial relationship building, cross-border settlements, simplifying complex reconciliations, new agent-vendor models, instant payments to reduce borrowing costs, and small loans.

In these scenarios, stablecoins are a better payment choice—and crucially, building the next generation on programmable money is easier than on traditional infrastructure. New relationships built on stablecoins will evolve into old relationships still using stablecoins. Over time, as full stablecoin payment platforms launch, stablecoins (which are cheaper, faster, and global) are likely to take a larger share of the payment mix.

Opportunities for New Payment Technologies

To understand what’s next, we should focus on technologies best suited for growing use cases.

Stablecoins—highly liquid assets backed 1:1, faster, cheaper, and global—are a new platform capable of serving underserved business categories today, like international and streaming payments. Key stablecoins are programmable, with features like arbitration, monthly (or hourly) billing, credit, escrow, and conditional payments, which can be flexibly extended to many new use cases. Unlike bank or card payments, stablecoin payments can be easily integrated into APIs, databases, and agent settlement flows, greatly simplifying reconciliation, approval, and onboarding—huge advantages for entrepreneurs building agent businesses.

Practically, stablecoins solve the unit economics issues of credit cards in extreme cases. No $0.30 minimum fee makes micro-payments feasible. No fees eating into large transfers means an agent paying $0.001 per second to a compute provider or a manufacturer settling a $50,000 vendor invoice can use the same rails. This flexibility is critical for engineers and entrepreneurs designing the next platform.

Building More Stablecoin Infrastructure

The most common objection to stablecoins is high deposit and withdrawal costs. For “tourists” unfamiliar with the system, this is true. But when a “tour guide”—the agent—assists, the problem disappears. The guide can help tourists exchange currencies and facilitate exactly the needed transactions, saving on fees.

Adding billing and arbitration features to our stablecoin “tour guide” system gets us close to the ideal.

Imagine shopping at Dell department store. You browse multiple brands, pick your items, and check out at a single register. The store handles the complex accounting across brands. Similarly, agents need a unified view of pending purchases across multiple vendors, with one-click approval for the entire batch. Users see “Your agent wants to book flights, hotels, and car rentals”—not three separate checkouts. The agent platform manages vendor relationships, while users manage intent. Users can approve, review, or dispute transactions.

Credit cards handle arbitration well, but new rails will need this layer too. When items are high-value or easily returnable, arbitration is easiest. Flights within a 24-hour cancellation window, unstarted subscriptions, profitable luxury goods—vendors can tolerate cancellations. But early-stage agent scenarios often involve low-margin digital goods like compute or API calls, or food delivery.

Summary

Smart agents won’t pay like tourists. They will pay like locals—through relationships, credit, and repeat transactions. This means real payment volume will flow through pre-negotiated B2B terms, not credit card swipes. Frankly, pre-negotiated B2B terms don’t require new payment rails. Settlement layers can be anything—wire transfers, ACH, or simple batch transfers. For established relationships, traditional payment methods work well.

But we are at a crossroads. Agents are emerging, entrepreneurs are building, and they need payment methods that work today—not waiting for years of credit card tech upgrades. Credit cards aren’t ready: too expensive for micro-payments, too challenging for reconciliation, burdened by technical debt, and requiring manual fraud decisions. Stablecoins are ready. They are programmable, global, easy to reconcile with digital services, and can be easily integrated into APIs and agent settlement flows. Even without negotiated merchant agreements or complex B2B terms, they can work from day one.

This is the opportunity window. Entrepreneurs building today will seek tools that work well now. Payments are sticky. Ultimately, new relationships built on stablecoins will evolve into old relationships still based on stablecoins. Over the next few years, as ecosystems mature, deposit friction will fade, and infrastructure gaps—billing, arbitration, credit, batch approval, interoperability—will be filled by startups building on stronger foundations.