Editor’s Note: This article offers a relatively grounded builder’s perspective: Over the past year, agent 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 e-commerce have gained increasing traction. However, after actively building products and engaging with merchants and developers, the author found that real demand has not yet emerged at scale.

The article breaks down several typical scenarios: agent-based shopping does not outperform traditional e-commerce in most categories, as users still require images, comparisons, and browsing; machine API payments seem suitable for stablecoin micropayments, but most developers today have already resolved payment needs through subscriptions, credit top-ups, and existing billing systems; payments between agents remain a long-term vision and are still in early stages, lacking real transaction volume.

Relatively speaking, agent-based finance is one of the few areas with existing demand. Funds, treasury teams, and DeFi users are already paying for financial tools, and AI can deliver tangible improvements such as real-time monitoring and automated rebalancing. However, this market also favors traditional institutions that already hold licenses, comply with regulations, and have established client relationships.

The author’s final judgment is that what the agent economy truly lacks is not merely 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 component. For giants, early positioning is a defensive move; but for startups, what truly matters is identifying markets that already exist today.

The following is the original text:

Over the past year, I’ve been building infrastructure for the agent economy and have engaged with teams at Stripe, Visa, Coinbase, Google, and dozens of startups advancing commercial applications of agents. I’ve mapped the landscape, launched a product, and sought to identify the real market.

But the reality is: genuine demand has not yet emerged. For startups looking to enter this space, numerous structural challenges remain.

Last month, Stripe launched 288 new products at the Sessions conference, and access to Agent-related documentation has approached 40% of total documentation views. Its Agent marketplace has already onboarded over 1,000 merchants. However, at the Sessions event itself, only a handful of Agents actually registered and completed transactions.

Visa noted that its Agent token currently requires a KYC approval process lasting 3 to 9 months, and companies must have annual revenues of at least $250 million to qualify for integration. Today, only companies of the caliber of Amazon and Walmart can successfully close the identity verification chain.

Coinbase previously reported that, as of April, there were 69,000 active agents and 165 million transactions on x402. However, independent on-chain analysis shows that the actual daily trading volume is approximately $17,000, with about half of those being test transactions (CoinDesk, March 2026).

What we learned while building shop.fast.xyz

We built shop.fast.xyz to directly validate agency-based commerce—real products, real merchants, real transactions.

However, for most product categories, the current AI shopping experience is clearly inferior to traditional e-commerce. When buying clothing, electronics, or furniture, users want to see images, browse options, and compare items side by side. A chatbot-style conversation is a step backward: it replaces a rich visual interface with a string of text. Human shopping begins with the eyes.

The agent performed well in what we originally thought was the most challenging part—it understands what users want and handles requests like “something similar but cheaper” effectively. The model layer works. However, it cannot replace the experience of simultaneously viewing ten products and choosing one. While the chat interface could incorporate product carousels and interactive displays, at that point, you’re essentially rebuilding an e-commerce frontend within the chat window. For shopping scenarios that require visual comparison, we haven’t yet found a compelling reason why a chat interface would be superior to the original e-commerce interface.

We do see demand from merchants, but this demand is largely defensive. Merchants want their stores to be discoverable by agents not because many consumers are currently shopping through agents, but because they fear being left behind if agents become the dominant channel in the future. This represents an opportunity in Agentic Engine Optimization, but it remains a “nice-to-have” rather than a “must-have.” Merchants are preparing in advance for a wave that has not yet arrived.

The areas where conversational commerce can truly enhance the experience are high-frequency, low-decision-cost purchase scenarios where users already know exactly what they want. The clearest example is ordering food: the market is large, the frequency is high, and decisions are made quickly—such as, “Order me a pad thai from my favorite place I ordered from last time.” In these scenarios, a conversational agent might outperform traditional interfaces. However, major food delivery platforms do not offer APIs. The only viable path is computer use—having the AI interact with apps visually, just like a human would. This process is slow, fragile, and the reasoning cost is simply unjustifiable for a $15 lunch.

Another opportunity lies in online stores so complex that users experience real frustration—such as layered discounts, coupon codes, membership points, and chaotic checkout processes. An agent that understands “help me apply my coupon, redeem my points, find the cheapest shipping option, and complete this in my language” can truly simplify today’s broken shopping experience. This is especially important for elderly users, non-native speakers, and those shopping across regions; or in highly specific scenarios where users have extremely niche, complex needs.

But both opportunities require massive B2C distribution capabilities. You're competing with DoorDash and Amazon for user access. Distribution scale at the consumer level is an advantage held by existing giants. The supply side of agency-based commerce is ready, but demand is constrained by user experience and distribution channels—more infrastructure cannot solve these two issues.

What did we learn from x402 and MPP?

We spoke with dozens of developers about their real payment needs. The pattern was nearly identical: current Agent API usage is essentially recurring consumption, such as for compute power, inference, and data sources. Developers already have subscription models, API keys, linked accounts, and billing relationships with core service providers.

The typical argument for stablecoin payments is that the effective minimum cost for credit card payments on Stripe is approximately 2.9% plus 30 cents, making API calls under $1 economically unviable. However, with today’s low transaction volumes, topping up credits resolves the issue—developers pre-load their accounts, eliminating the problem.

A deeper issue lies in the vendor market: most large SaaS companies do not want to offer fragmented API access at fractions of a cent. Their business model is based on multi-year enterprise contracts. Companies that rely on substantial committed revenue resist new pricing models that bypass this structure.

The machine commerce ecosystem is structurally a long-tail market, serving small-scale services, vertical data sources, independent developers, MCP servers, and more. Protocols like MPP and x402 are well-suited to this niche. However, by definition, this is a market catering to specialized needs—and developers have historically been among the least willing to pay.

When Stripe Projects launched, it integrated with 32 service partners, including Vercel, Supabase, Cloudflare, and Twilio, covering most core services developers use to build and deploy software—all accessible through the existing billing system. The top of the developer tech stack is already well-served. The opportunity in the new payment rail lies in everything beyond the top 30 providers: it exists, but its scale is naturally smaller than the market space suggested by those grand narratives.

The same logic applies to content access. Agents are continuously scraping and summarizing articles, and publishers are beginning to push back. However, when content monetization truly scales up, it is likely to be facilitated through existing CDN providers positioned between publishers and the internet—such as Cloudflare, which has already launched AI auditing tools—or via bulk licensing agreements between publishers and AI labs. Infrastructure opportunities will flow to established players who already possess distribution capabilities.

What we learned from Agent-to-Agent payments

The commerce between agents is a long-term vision, but it currently remains almost entirely theoretical. No one has yet achieved meaningful trading volume. The truly challenging aspects are being advanced by various startups, including agent discovery, trust establishment, term negotiation, and dispute resolution.

Once this transaction structure truly takes shape, it will look entirely different from existing payment rails. Neither party will have a human identity; settlement requirements will be under one second; transaction amounts can range from fractions of a cent to millions of dollars; and it will involve multi-party settlement rather than the default bilateral buyer-seller model of current payment systems. When it finally happens, we believe it will explode at an unprecedented speed and scale.

This is precisely a long-term bet on dedicated settlement infrastructure—and it’s a real one. But a “real long-term bet” is not the same as the “current market.” We were once among those who claimed for months on end that this market was coming, and over the past few years we’ve built an entire infrastructure around it, including our distributed network. Theoretically, it can scale to over one billion TPS with latency under 50 milliseconds and an average consensus time of 10 milliseconds. But we must return to where the market actually is today.

What we learned in Agent Finance

It can be said that this is the only category with existing real demand. Customers already exist and are already paying. Fund managers, asset management teams, and DeFi users are already spending money on financial tools today. Integrating AI into existing workflows is a natural product path.

Agents will also create entirely new behavioral patterns. Agents capable of autonomously monitoring and rebalancing hundreds of positions in real time operate in ways that humans cannot manually replicate. This represents genuine augmentation, not just automation.

The challenge lies in the competitive landscape. The financial industry is highly regulated and relies heavily on established relationships. Existing institutions possess licenses, compliance infrastructure, and customer relationships. Startups can enter through less regulated areas, such as DeFi, or target segments where incumbents move slowly, or leverage AI to create new capabilities that giants currently lack. However, overall, the competitive dynamics in this space favor established players more than in the previous three categories, because adding AI to existing products and customers is far easier than starting with AI and then trying to build products and customer bases from scratch.

Honest summary

So why are people still doing this? There are two reasons.

The first is the incentive mechanism. Large companies have sufficient cash flow to bet on a future that may take years to materialize. For them, the cost of entering five years early is merely a rounding error; but delaying entry by even one year could be catastrophic. So they must act.

The second is a cognitive blind spot. When your business is payments, every problem looks like a payment problem. Since the agent economy needs a payment layer, everyone builds a payment layer.

But payment is only one part of a larger problem. The real challenge isn't getting money to flow between agents—it’s how to coordinate work between agents and humans, how to verify that tasks have been completed, and how to settle outcomes. Payment is just one part of settlement. Settlement is just one part of coordination. And coordination is the real prize.

Large-scale coordination naturally creates a demand for settlement mechanisms. Payments will become one instrument in this symphony, not the entire composition. Companies that truly solve coordination problems will ultimately incorporate payments into their solutions, rather than having payment companies absorb coordination.

Most existing giants are defensively building a future of "machine-scale trading." For them, the timeline isn't critical, as they have near-infinite runway.

But startups don’t have that luxury. We must find where the market truly is right now. We can’t keep waiting for the wave to arrive.

A year of development has led us in an unexpected direction. There is indeed activity there, and it is growing rapidly and underserved. It exists outside the four categories we have identified.