Agent支付這一年：光環之下，市場未至

Original Title: a year inside agentic payments: the uncomfortable truth

Original Author: @13yearoldvc

Original Translation: Peggy

Editor's Note: This article offers a relatively calm builder's perspective: Over the past year, agentic payments have become a hot narrative at the intersection of AI, payments, and crypto, with companies like Stripe, Visa, Coinbase, and Google all making moves. Concepts such as stablecoin micropayments, x402, machine-to-machine settlement, and agent-powered e-commerce are heating up. However, after actually building products and engaging with merchants and developers, the author found that real demand has not yet materialized on a large scale.

The article breaks down several typical scenarios: Agent-driven shopping is not superior to traditional e-commerce for most product categories, as users still need images, comparisons, and browsing. Machine API payments seem suitable for stablecoin micropayments, but most developers currently solve this through subscriptions, prepaid credits, and existing billing systems. While payments between agents are a long-term vision, they remain in the early stages with a lack of real transaction volume.

Relatively speaking, Agent Finance is one of the few areas with existing demand. Funds, treasury teams, and DeFi users already pay for financial tools, and AI can provide tangible capability improvements like real-time monitoring and automated portfolio rebalancing. However, this market also favors traditional institutions that already possess licenses, compliance, and customer relationships.

The author's final judgment is that what the agent economy truly lacks is not just a payment layer, but more complex coordination capabilities – how to enable agents to collaborate with humans, verify task completion, and settle outcomes. Payment is just one piece of the puzzle. For giants, positioning early is a defensive choice; but for startups, what truly matters is finding a market that already exists today.

Below is the original text:

Over the past year, I've been building infrastructure for the Agent economy and have spoken with teams at Stripe, Visa, Coinbase, Google, and dozens of startups advancing agent commerce. I've mapped out the landscape, launched products, and tried to find the real market.

But the reality is: real demand hasn't arrived yet. For startups looking to enter this space, there are still many structural problems.

Stripe launched 288 new products at its Sessions conference last month, and Agent-related documentation now accounts for nearly 40% of all document views. Its Agent Commerce marketplace has onboarded over 1,000 merchants. Yet, at Sessions, the number of registered agents that actually completed transactions was in the single digits.

Visa mentioned that its Agent token currently requires 3 to 9 months of KYC approval, and generally requires companies to have at least $250 million in annual revenue to be eligible. Today, only companies the size of Amazon or Walmart have the capacity to close the identity verification loop.

Coinbase previously reported 69,000 active agents and 165 million transactions on x402 as of April. But independent on-chain analysis shows real daily transaction volume is around $17,000, with about half being test transactions (CoinDesk, March 2026).

What We Learned Building shop.fast.xyz

Agent to Merchant: Agentic Commerce

We built shop.fast.xyz to directly validate agentic commerce. Real goods, real merchants, real transactions.

But for most product categories, the current AI shopping experience is clearly inferior to traditional e-commerce. When buying clothes, electronics, or furniture, users want to see images, browse options, and compare side-by-side. A chatbot-style conversation is actually a regression: you replace a rich visual interface with a string of text. Human shopping is, first and foremost, visual shopping.

The agent performs well in areas we initially thought would be hardest. It understands what the user wants and handles requests like "similar to this, but cheaper." The model layer works. But it can't replace the experience of looking at ten products simultaneously and choosing one. You can add product carousels and interactive displays to the chat interface, but at that point, you're essentially rebuilding an e-commerce frontend within a chat window. For shopping scenarios requiring visual comparison, we haven't found a convincing answer for why the chat shell is better than the original e-commerce interface.

We do see demand on the merchant side, but it's more defensive. Merchants want their stores to be discoverable by agents, not because many consumers are shopping via agents today, but because they fear being left behind if agents become a mainstream channel. This is the Agentic Engine Optimization (AEO) opportunity, but it's currently a "nice-to-have," not a "must-have." Merchants are preparing for a wave that hasn't arrived yet.

Where conversational commerce can truly improve the experience is in high-frequency, low-decision-cost purchases where users already know what they want. The clearest example is ordering food. The market is large enough, the frequency is high enough, and the decision is fast enough, like "order Pad Thai from the place I liked last time." In this scenario, a conversational agent might win. But major food delivery platforms don't open their APIs. The only path is computer use, where AI visually operates apps like a human. This process is slow, fragile, and the inference cost doesn't make sense for a $15 lunch.

Another opportunity is complex online stores that genuinely frustrate users. Think stacks of discounts, promo codes, loyalty points, and messy checkout flows. An agent that can understand "apply my coupon, use my points, find the cheapest shipping, and complete the purchase in my language" could simplify a currently broken shopping experience. This is especially relevant for elderly users, non-native speakers, cross-regional shopping, or very specific scenarios with niche, complex needs.

But both opportunities require massive B2C distribution capabilities. You're competing with DoorDash and Amazon for user entry points. Consumer-scale distribution is the advantage of existing giants. The supply side of agentic commerce is ready, but the demand side is limited by user experience and distribution channels, and more infrastructure won't solve these two problems.

What We Learned from x402 and MPP

Agent to Web/API: Machine Commerce

We spoke with dozens of developers about their real payment needs. The pattern is almost identical: today's Agent API usage is essentially recurring consumption, for things like compute, inference, and data sources. Developers already have subscriptions, API keys, linked accounts, and billing relationships with core service providers.

The typical argument for stablecoin payments is that credit card payments on Stripe have an effective minimum cost of around 2.9% plus $0.30, making API calls under $1 uneconomical. But at today's low transaction volumes, prepaid credits solve the problem. Developers pre-fund their accounts, and the issue disappears.

The deeper issue is the supplier market. Most large SaaS companies don't want to offer granular API access for fractions of a cent. Their business models rely on multi-year enterprise contracts. Companies dependent on large committed revenue will resist new pricing models that bypass this structure.

Structurally, machine commerce is a long-tail market. It serves small services, niche data sources, independent developers, and MCP servers. Protocols like MPP and x402 are well-suited for this segment. But by definition, this is a market for specialized, power users; and developers have historically been among the least willing to pay.

When Stripe Projects launched, it onboarded 32 service provider partners, including Vercel, Supabase, Cloudflare, Twilio, etc., covering most core services developers use to build and deploy software, all accessible through existing billing systems. The top of the developer technology stack is already well-served. The opportunity for new payment rails lies outside those top 30 providers: it's real, but its scale is inherently smaller than the grand narratives suggest.

Content access follows the same logic. Agents are already scraping and summarizing articles, and publishers are starting to fight back. But when content monetization truly scales, it will likely happen through CDN providers that already sit between publishers and the internet, like Cloudflare with its AI audit tools, or through bulk licensing agreements between publishers and AI labs. Infrastructure opportunities will flow to existing players with distribution.

What We Learned from Agent-to-Agent Payments

Commerce between agents is the long-term vision, but currently, it's almost entirely theoretical. No one has generated meaningful transaction volume yet. The truly difficult parts are being tackled by various startups, including agent discovery, trust establishment, terms negotiation, and dispute resolution.

Once this transaction structure truly takes shape, it will look completely different from existing payment rails. Both parties lack human identity; latency requirements are sub-second; transaction amounts can range from fractions of a cent to millions of dollars; and it will involve multi-party settlement, not the default bilateral buyer-seller model of existing rails. When it happens, we believe it will explode with incredible speed and scale.

This is the long-term bet dedicated settlement infrastructure, and it's a real bet. But "a real long-term bet" is not the same as "the current market." We were among those who proclaimed this market was coming for months, building a whole infrastructure stack around it over the past few years, including our distributed network. Theoretically, it can scale to over 1 billion TPS with sub-50ms latency and 10ms average consistency. But we must return to where the market is right now.

What We Learned from Agent Finance

This is arguably the only category with existing real demand. Customers exist and are already paying. Fund managers, treasury teams, and DeFi users are already spending money on financial tools. Plugging AI into existing workflows is a natural product path.

Agent Finance will also create entirely new behaviors. An agent that can autonomously monitor and rebalance hundreds of positions in real-time can operate in ways humans cannot manually replicate. There is genuine capability improvement here, not just automation.

The challenge is the competitive landscape. The financial industry is highly regulated and relationship-driven. Incumbents possess licenses, compliance infrastructure, and customer relationships. Startups can enter in less regulated areas, like DeFi; or find areas where incumbents move slowly, or where AI creates new capabilities that giants lack. But overall, competitive dynamics in this category favor incumbents more than the previous three, because adding AI to existing products and customers is far easier than starting from AI and building products and customers.

An Honest Summary

So, why is everyone still doing this? Two reasons.

First, incentives. Large companies have enough cash flow to bet on a future that may take years to materialize. For them, entering five years early is a rounding error; entering one year late could be catastrophic. So they must do it.

Second, cognitive bias. When your business is payments, every problem looks like a payments problem. The Agent economy needs a payment layer, so everyone builds a payment layer.

But payment is just one part of a larger problem. The truly hard problems are not moving money between agents, but how to coordinate work between agents and humans, how to verify task completion, and how to settle outcomes. Payment is just a component of settlement. Settlement is just a component of coordination. And coordination is the real prize.

Large-scale coordination will naturally create a need for settlement mechanisms. Payment will become one instrument in this orchestration, not the entire composition. Companies that truly solve coordination will ultimately incorporate payment, not the other way around.

Most existing giants are defensively building for a future of massive machine-to-machine transactions. For them, the timeline doesn't matter, as they have nearly infinite runways.

But startups don't have this luxury. We must find where the market actually is right now. We can't just wait for the wave to hit.

A year of building has led us in an unexpected direction. There is activity there, growing fast and underserved. It exists outside the four categories we mapped out.

Original Link