Source: Tourists in the bazaar: Why agents will need B2B payments — and why stablecoins will get there first
>
Author: Sam Broner, a16z; Translation: Jinse Finance
>
Walking into the bazaar as a tourist, you'll see a bustling scene: crowds thronging around, eyeing goods, comparing prices, trying them out, haggling with each vendor, and paying. It all seems like one-off transactions—each interaction a small negotiation, trust maintained by cash or credit cards.
>
But the real business in the bazaar doesn't work that way. Look closer: most people are locals, heading directly to familiar merchants. Restaurant owners seek out familiar butchers, fishmongers, and farmers; tailors go to mechanics, weavers, and artisans. They all pay with credit.
When we discuss how AI agents will handle payments, we subconsciously think like tourists. But AI agents will act like locals. The characteristics that distinguish agents from humans—unlimited replication, flexible resource allocation, and zero startup costs—mean that a few agents can dominate a niche market. Even as building an agent network becomes easier, relationships, partnerships, and trust can still create a winning experience. Top agents don't need the payment channels tourists use; they need merchant relationships, operating capital, and credit lines. Agents can guide tourists (like you) through a marketplace. What will this look like? When agents are integrated into quasi-business platforms, agent payments must shift from retail payment channels to pre-negotiated B2B terms and credit, a need that existing payment systems cannot fully meet. This is the opportunity for next-generation payment channels (like stablecoins)—provided entrepreneurs can create excellent solutions for next-generation payment scenarios such as agents, streaming payments, and high-frequency, low-value global transactions. This article explores this viewpoint in three parts: 1. How do agents differ from humans, and how do these differences determine winning payment strategies? 2. Why are existing solutions insufficient? 3. What needs to be built for next-generation payment channels to succeed? I. Differences Between Agents and Humans To understand agents and payments, two questions must be answered: Do agents behave like humans or businesses? Do agents play the long game or the short game? The answer is: Agents behave more like businesses, building long-term relationships with suppliers and partners. They are lightly customized instances on top of large business structures—like top tour guides in well-connected travel agencies, or franchisees who don't renegotiate the supply chain but only fine-tune solutions to local tastes. Why do agents operate like businesses? First, the best experience is well-designed. I don't want an agent haggling with merchants, comparing prices, and only discussing terms at checkout. I want an agent who can handle all of this in advance: knows which companies are reliable, has pre-agreed prices, and allows for one-click payment. This is a business relationship, not a touristy, one-off transaction. In fact, human agents have existed for a long time: travel agents, copyright agents, artist management, watch dealers, real estate agents, etc. They build crucial, multi-round relationships with publishers, production companies, watch dealers, and lending institutions, customizing each transaction based on these relationships. Second, agents can be infinitely replicated, but scalable commerce (and its advantages) cannot. The best agents leverage the costs and benefits of scale: cheaper computing power, better merchant pricing, deeper system integration, and more deterministic components. Scale begets scale. A travel agent booking 1 million tickets a year will certainly get better airline terms than one booking only 10. We've already seen this trend: only giants like ChatGPT have the channels to partner with Shopify, Amazon, and E-commerce. Small startups can only use automated browsers or reverse APIs, and they still have to pay retail fees. This is why agencies are becoming centralized, or rather, most agencies are built on larger platforms. Building an agency network is easy, but economic principles dictate that only a few agencies will remain in each vertical market – each with deep merchant relationships and profits reinvested in optimizing the user experience. Vertical agencies with deep merchant relationships can work in conjunction with user agencies, achieving a win-win situation. Two Types of Payment Relationships If agencies operate like businesses, two types of payment relationships need to be designed: User → Agent; Agent / Agent Platform / Agent "Guide" → Merchant. Users pay agents: subscriptions, pay-per-use, credit lines, or authorized access to user accounts. Agents pay merchants: negotiated B2B terms, bulk pricing, 30-day payment terms, or payments to lower-level agents. Based on existing business expenditures, agents may occasionally use retail channels to pay merchants, but even then, it only accounts for a small portion of total expenditure. This is essentially the operating logic of credit cards today: the issuing bank and the consumer have a retail relationship, with the issuing bank bearing the risk, designing customized rewards, and providing credit; the acquiring bank and the merchant have a commercial relationship, with negotiated terms, large-scale transfers, and complex operational funding arrangements. Agents and Credit Cards: Seemingly a Perfect Match, But Difficult to Implement Many people say that credit cards are actually quite suitable for agent scenarios: wide acceptance, a friendly range of $20–$1000, built-in arbitration, cancellation mechanisms, and digitalization. Credit cards also have monthly statements—a key entry point for users to understand their spending. With agents replacing children who spend recklessly as the main cause of unexpected expenses, this function will certainly be iterated upon repeatedly. However, credit cards have two fatal problems: technically unsuitable for agents; and the fee model puts card organizations in the classic innovator's dilemma. Card Payment Technology is Difficult to Upgrade Almost all card payment technologies assume a human-in-the-loop: approver, UI interface, traditional payment types (one-time, subscription). Dozens of card virtualization products, such as Stripe Link and Visa 3D—allowing you to store your cards on websites and subscribe to automatic payments—are finally working, but this technology took over 15 years to mature. The rapid adoption of agents, with thousands of payment service providers, POS machines, merchants, and clients, simply cannot keep up with the slow upgrades to interfaces, programmability, and risk control systems for this new payment flow. Card payments are completely ineffective in extremely high/low amount scenarios. Imagine: agents paying streaming providers or making micropayments for API calls. These don't work on card channels: Visa doesn't support payments less than 1 cent; its economic model defaults to a fixed fee of about 30 cents. Visa isn't incapable of developing streaming or micropayment technology, but getting stakeholders accustomed to lower revenue is extremely difficult. Even more problematic is that card organizations are caught in the innovator's dilemma. Even with similar user relationships and needs, proxy payments often fall outside the comfort zone of $20–$1000. Worse still, many early scenarios involve payment APIs—services that are difficult to refund and easily resold (highly susceptible to fraud). Credit cards aren't unusable, but history has repeatedly shown that the innovator's dilemma ultimately weakens giants. Even setting aside bank cards, traditional payments still have a place in the future. The Positioning of Existing Payments & Opportunities for Stablecoins When proxies integrate into quasi-business platforms, most high-value expenditures will shift to pre-negotiated B2B terms: invoices, 30-day payment terms, discounts, and credit lines. In this model, the "payment channel" can be anything—usually the tedious asynchronous settlement on traditional channels. Fees are spread across large transactions, and operating capital can be negotiated between businesses. But proxies won't just survive in this mature system. Agents are emerging and are active in areas where traditional payments are not well-suited: Initial Partnerships, Cross-Border Settlements, Simplified Complex Reconciliation, A New Agent-Merchant Model, Instant Payments to Reduce Borrowing Costs, Microloans. In these scenarios, stablecoins are a better choice. Crucially, building next-generation functionality on programmable money is far easier than on traditional infrastructure. New relationships established with stablecoins will continue to use stablecoins once they mature. As complete stablecoin payment platforms launch, faster, cheaper, and more global stablecoins are likely to occupy an increasingly larger share of the payment structure. The Opportunity Window for New Payment Technologies: The technology best suited for agent scenarios is one that can match the growing demand. Stablecoins—backed 1:1 by high-quality liquid assets, faster, cheaper, and global—are a brand-new platform that meets the needs of currently neglected business categories, such as international payments and streaming payments. Crucially, stablecoins are programmable. Core functions such as arbitration, monthly (or hourly) billing, credit, custody, and conditional payments can be flexibly expanded to support numerous new scenarios. Unlike bank or card payments, stablecoins can be integrated extremely easily into APIs, databases, and agent checkout processes, significantly simplifying reconciliation, approvals, and account opening—a huge advantage for entrepreneurs eager to build agent businesses. From a practical cost perspective: stablecoins solve the unit economics problem of credit cards at extreme amounts. The absence of a 30-cent minimum fee makes micropayments possible; large transfers are free from card organization fees eroding profits. An agent paying $0.001 per second to a computing power provider and a manufacturer settling $50,000 in supplier invoices can use the same payment channel. This flexibility is crucial for engineers and entrepreneurs choosing next-generation platforms. More stablecoin infrastructure must be built. The most common concern about stablecoins is the high cost of fiat currency deposits and withdrawals. This is true for unsuspecting users, but this problem disappears if users are accompanied by a "guide"—or agent. A guide can help users exchange currency, accurately complete necessary transactions, and save on fees. Adding billing and arbitration functions to stablecoin-supported "guides" brings us closer to the system we need. Imagine you're shopping at Bloomingdale's: you visit multiple counters, accumulate a bunch of items, and finally check out all at once. The mall distributes the money to each merchant. Agents need a similar model: a unified purchasing view across multiple merchants, with one-click batch approval. Users see: "Your agent wants to book flights, hotels, and car rentals," instead of three separate checkout processes. The agent platform handles merchant relationships; users are only responsible for intent. Users can approve, view, or challenge transactions. Credit card arbitration mechanisms are well-established, but new channels need to add this layer. High-margin, easily returned goods are easiest to arbitrate: cancelable airline tickets within 24 hours, unstarted subscriptions, high-margin luxury goods—merchants can afford refunds. But early agency scenarios often involve low-margin digital goods: computing power, API calls, food delivery, etc. Conclusion: Agents won't pay like tourists. They'll pay like locals—based on relationships, credit, and repeat purchases. This means that real payment volumes will follow pre-negotiated B2B terms, not credit card swipes. Frankly, mature B2B terms don't require new payment channels. The settlement layer can be anything: wire transfer, ACH, bulk transfers. Traditional payments are perfectly adequate for mature relationships. But we are at a crossroads. Agents are booming, and entrepreneurs are building them. They need usable payments now, not years to upgrade the card ecosystem. Credit cards aren't ready: micropayments are too expensive; reconciliation is too difficult; they're burdened by technical debt; and they rely on "human-in-the-loop" fraud decisions. Stablecoins are ready. Programmable, global, easy to reconcile with digital services, and easily integrated into APIs and agent checkout processes. Even without negotiated merchant agreements or complex B2B terms, stablecoins are usable from day one. This is the window of opportunity. Entrepreneurs building agents today will prioritize tools that are usable now. Payments have strong stickiness. Ultimately, new relationships built on stablecoins will remain on stablecoins once they mature. In the next few years, the ecosystem will mature, deposit and withdrawal frictions will disappear, and infrastructure gaps in billing, arbitration, credit, batch approval, and interoperability will be filled by a wave of startups built on a more robust underlying layer.
JinseFinance
Alex
Cointelegraph