Author: jessy
Translated by: Jiahuan, ChainCatcher
For the past year, I've been working on building infrastructure for the Agent economy, having conversations with teams at Stripe, Visa, Coinbase, Google, and dozens of startups pushing Agent commerce. I mapped the entire industry, shipped products, and tried to find product-market fit.
There is no real demand yet, and startups face multiple structural problems when entering this space.
Last month, Stripe launched 288 new products at Sessions, with their Agent documentation nearing 40% of total documentation readership. Their Agent commerce marketplace had over 1,000 enabled merchants. Yet, the number of registered Agents that transacted at Sessions was in the single digits.
Visa mentioned that their Agent payment tokens (tokenized payment credentials bound to an Agent for making payments on a user's behalf) currently require 3 to 9 months for KYC approval, and effectively need a minimum revenue threshold of $250 million to qualify. Today, only companies on the level of Amazon or Walmart can close this identity loop.
Coinbase reported 69,000 active Agents and 165 million transactions on the x402 protocol as of April. But independent on-chain analysis shows an actual daily transaction volume of about $17,000, with roughly half being test transactions (according to CoinDesk, March 2026).
Agent to Merchant
We built shop.fast.xyz to directly validate the practical application of concierge-style commerce. It included real products, real merchants, and real transactions.
For most product categories, the current AI shopping user experience is entirely inferior to traditional e-commerce. When you buy clothes, electronics, or furniture, you want to see pictures, browse options, and compare.
The conversational chatbot format is actually a step backwards. You're replacing a rich visual interface with pure text conversation, and humans are fundamentally visual shoppers.
Agents excel in areas we thought would be difficult. They can understand user needs and properly handle instructions like "something like this but cheaper." The model layer works.
But it cannot replace the experience of browsing ten products side by side and picking one. Chat interfaces can be enhanced with carousels and interactive displays, but at that point, you're essentially rebuilding an e-commerce frontend inside a chat window. For visual-driven comparison shopping, we haven't found a compelling reason why a chat interface is better than a native e-commerce interface.
We saw real demand from merchants, but it's a defensive demand.
Merchants want their stores to be queriable by Agents. Not because customers are buying via Agents now, but because they fear being left behind if this becomes the mainstream channel.
It's an "Agent Engine Optimization (AEO)" strategy, but currently it's just a nice-to-have, not a must-have. Merchants are preparing for a wave that hasn't arrived yet.
Conversational commerce does improve the experience in certain scenarios: high-frequency, low-decision-cost purchases where the user already knows what they want. Ordering food delivery is the clearest example. Large market, extremely high frequency, rapid decisions ("order me pad thai from that place I used last time"). Conversational Agents have a chance here.
But the major food delivery platforms haven't opened their APIs. The only route is "computer use": having AI visually navigate and operate the app like a human. This is slow, brittle, and the inference costs simply don't pencil out for a $15 lunch order.
Another breakthrough point lies where: certain shops have excruciatingly complex UI navigation. Layers of discounts, promo codes, loyalty programs, and confusing checkout flows.
An Agent that understands "use my coupon, deduct my rewards points, find the cheapest shipping, operate in my native language" can simplify parts of the internet that are terrible today. This is significant for older users, non-native speakers shopping at foreign online stores, or highly specific scenarios with very niche needs.
Both of these breakthrough points require massive consumer-facing (B2C) distribution. You're competing with DoorDash (the largest food delivery platform in the US, holding 56% market share) and Amazon for the user entry point.
Consumer-scale distribution is an incumbent advantage. The supply side for concierge commerce is ready, while the demand side is constrained by user experience and distribution channels. Building more infrastructure does not solve these two problems.
Agent to API
We spoke with dozens of developers about their actual payment needs. The picture is strikingly consistent: Agent-to-API usage today is routine for compute, inference, and data sources. Developers already have subscription services, archived API keys, and billing relationships with core vendors.
The typical stablecoin argument is: on Stripe, the minimum effective cost for credit card processing is around 2.9% plus 30 cents, making sub-dollar API calls uneconomical. But for the current low transaction volume, prepaid credits solve this. Developers top up their accounts in advance, and the problem disappears.
The deeper problem is the vendor marketplace. Most mainstream SaaS companies do not want to offer fractional-cent ephemeral API access. Their business model is multi-year enterprise contracts. Companies whose revenue relies on large commitment deals resist pricing mechanisms that circumvent their existing model.
Machine commerce is structurally a long-tail market, including smaller services, niche data sources, individual developers, and MCP servers. Protocols like MPP and x402 are well-suited for this segment.
But by definition, this is a market serving power users with special needs, and historically, developers have been one of the least willing-to-pay segments.
When Stripe Projects launched, it partnered with 32 vendor partners like Vercel, Supabase, Cloudflare, Twilio, etc., covering most tools developers use to build and deploy software, all accessible via existing billing systems. The top-of-stack demand for the developer tech stack is already met.
The opportunity for new payment rails exists for everything beyond those top 30 services: the opportunity is real, but its scale is inherently much smaller than the flashy numbers suggest.
The same dynamic applies to content acquisition. Agents are already scraping and summarizing articles, and publishers are fighting back.
But when content monetization happens at scale, it will come through CDN vendors already sitting between publishers and the internet (Cloudflare has already launched AI audit tools for this), or through large-scale licensing deals between publishers and AI labs.
This infrastructure opportunity will ultimately flow to incumbents who already own distribution channels.
Agent to Agent
Agent-to-Agent commerce is a long-term vision, currently almost entirely theoretical, with no one achieving meaningful transaction volume. Startups are tackling the core challenges: Agent discovery, trust establishment, terms negotiation, and dispute resolution.
When this transaction structure materializes, it will look fundamentally different from existing payment rails. Neither party in the transaction contains a human identity. Latency is sub-second. Funds ranging from fractions of a cent to millions of dollars move in the same flow.
There's also multi-party settlement, which doesn't fit the bilateral buyer-seller model existing payment rails assume. Once this happens, we believe it will come fast and big.
This is a long-term bet on dedicated settlement infrastructure, and it's real. But a "real long-term bet" is not the same as "current market."
For months, we were also evangelizing this market and built a full infrastructure around it over the past few years. With our distributed network, we could theoretically scale to over 1 billion TPS with <50ms latency and 10ms average consensus. But we have to meet the market where it actually is today.
Agent to Finance
This is arguably the only category with pre-existing demand. The customer base already exists and is willing to pay. Today, fund managers, finance teams, and DeFi users pay for financial tools. Embedding AI into existing workflows is a natural product evolution.
Agent finance also creates entirely new behaviors. An Agent that can autonomously monitor and rebalance hundreds of positions in real-time operates in ways a human cannot manually replicate. This is not just automation; it's a substantive capability increase.
The challenge is the competitive landscape. The financial industry is heavily regulated and relies on entrenched business relationships. Incumbents have licenses, compliance infrastructure, and client relationships. Startups can find a place in less regulated areas (like DeFi), where incumbents move slowly, or where AI creates capabilities incumbents don't have.
But compared to the other three categories, the competitive dynamics here are more favorable to incumbents, because layering AI on top of existing products and customer bases is far easier than doing the reverse.
The Real Game
So, why are people still building these things? Two reasons.
First, motivation. Industry giants have ample cash flow to bet on a future that may take years to materialize. For them, the cost of being five years early is a rounding error, while the cost of being one year late is existential. So they must build.
Second, cognitive bias. When your core business is payments, every problem looks like a payments problem. The Agent economy needs a payments layer? Build the payments layer.
But payments are just one part of a larger problem. The real challenge isn't how to move money between Agents; it's coordinating work between Agents and humans, verifying the work, and settling on the outcome. Payments are just part of settlement. Settlement is just part of coordination. And coordination is the real prize.
Coordination at scale will naturally spawn settlement mechanisms as a necessity. Payments are just an instrument in the symphony, not the entire score. Companies that solve coordination will subsume payments, not the other way around.
Most incumbents are building defensively for a future where machines transact at scale. Since their funding runway is infinite, the timeline doesn't matter to them.
But startups don't have that luxury. We must go find where the market actually is and can't wait for the wave to hit.
A year of building has led us to an unexpected direction. There is real market activity there, growing fast, and underserved. It sits outside the four categories we painted.
