Only 14% of consumers trust AI shopping recommendations. 86% verify every output manually. Yet Visa just committed billions to a platform that only works if agents buy. This is the definition of infrastructure before demand.
Let me explain why this matters. I’ve spent the last decade reading order books and on-chain flows. I’ve seen protocols build complex yield farms before anyone understood LPs. The pattern is always the same: smart money lays the pipes, retail pays for the water later. Visa’s Agent Score and Agentic Directory are exactly that—pipes for a market that doesn’t exist yet.
The Core Thesis Visa’s vision is simple: create a trust layer for autonomous agents. When an AI wants to buy something on behalf of a human, neither the merchant nor the consumer has reason to trust each other. The merchant fears a fraudulent agent. The consumer fears a compromised agent spending too much. Visa steps in with a centralized verification system—Agent Score rates the reliability of each agent, and the Agentic Directory maps which agents are allowed to transact on which merchant sites. They wrap this with tokenized credentials (replacing raw card numbers with single-use tokens) and a human-in-the-loop requirement for high-value purchases.
Technically, this is sound. It reuses Visa’s existing settlement rails, fraud detection models, and bank partnerships. Over 30 issuing banks in Europe have signed on. The stablecoin settlement volume hit a $70 billion annualized run rate. But here’s the catch: most of that volume is test transactions, not real commerce. The real agent-powered purchases are still measured in thousands, not billions.
The Data That Disturbs I ran the numbers from the Product.ai survey cited in Visa’s own strategy brief. 47% of US consumers have used an AI tool to shop. But only 14% trust the recommendations enough to let the agent execute without manual approval. That means 86% still babysit every transaction. If you remove the “research” phase and focus only on autonomous purchasing, the actual agent commerce market is <1% of total e-commerce. Visa is building a toll road for a village that doesn’t have cars yet.
This isn’t a dismissal of the vision. It’s a timing bet. In 2017, I wrote a Python script to scrape Ethereum mainnet for new ERC-20 tokens selling at ICO prices before the crowd saw the listings. I made 400% in weeks. But I got in early because I understood the technical arbitrage, not because the market was ready. Visa is doing the same—they’re buying the geometric advantage of being the first trust anchor in a world that will eventually need one. But the “eventually” could be 3 to 5 years away. And during those years, the infrastructure investment burns cash without proportional revenue.
The Contrarian Angle: Crypto-Native Rails Are the Real Threat Everyone is watching Visa vs Apple vs Google in the so-called “three-rail war” for agent payments. But the market is ignoring the dark horse: crypto-native payment channels like x402 and MPP. These protocols don’t require a trusted human to approve each transaction. Instead, the agent holds a crypto wallet with pre-set spending limits. It signs transactions on-chain. No middleman. No 2.5% merchant fee. No human verification overhead.
Visa’s “human-in-the-loop” is a feature now, but it’s a liability later. The whole point of an AI agent is autonomy. If you still need to click Approve, you’re just using a smarter search engine. Crypto rails remove that friction entirely. The agent buys, the wallet deducts, the merchant gets paid instantly in USDC or ETH. The only trust required is the smart contract logic and the agent’s private key management.
The problem is key management. If a user loses the agent’s private key, the funds are gone. Visa solves this with centralized recovery. But as AI wallets mature (account abstraction, social recovery, multi-sig agents), the crypto advantage grows. I’ve been tracking x402 usage since late 2025. It’s still niche—a few thousand transactions per month. But the growth curve is steep, and the unit economics beat Visa’s by a factor of 5 on fees.
Why the Market Is Wrong About Visa’s Moat The street sees Visa’s brand, regulatory compliance, and bank relationships as an unassailable moat. They’re wrong. Moats in technology are not about brand—they’re about switching costs and network effects. Visa’s switch cost for an agent commerce context is zero if the merchant integrates a crypto payment button alongside Visa. Merchants already accept crypto through services like BitPay. Adding x402 as a option is a weekend of code. Meanwhile, Visa’s Agent Score requires merchants to submit to a centralized rating system that Visa controls. Future updates to the score algorithm could deprioritize a merchant’s agents overnight. That’s not a feature—it’s a liability.
From my own experience negotiating institutional custodial solutions in 2024, I learned that institutions prioritize control over cost—until they don’t. When the first major hack of a centralized agent directory happens (and it will), the narrative flips overnight. Trust becomes distrust. Then the market pivots to trust-minimized architecture.
The Real Risk: Infrastructure Spend Becomes Sunk Cost Visa CTO confirmed the core problem: “How do consumers trust AI to act on their behalf?” The answer Visa built is “by having us watch every move.” That’s not a solution—it’s a crutch. If agents require human approval, they’re not agents—they’re assistants. The entire value proposition of autonomous commerce collapses. The $70 billion stablecoin settlement number includes a lot of noise. I estimate less than 5% of that comes from agent-initiated transactions. The rest is standard e-commerce settling in stablecoins.
This is the classic trap I saw in DeFi yield farming in 2020. Everyone chased high APY on unaudited pairs until one impermanent loss event wiped them out. The smart money rotated into stablecoin pairs. The laggards got rekt. Today, the smart money in agent commerce is building on unpermissioned rails that don’t require a gatekeeper. If I were deploying capital, I’d put it into protocols that give agents native wallets with programmable spending rules—not into Visa’s stock.
Takeaway: The Chop Is the Opportunity Sideways markets reward positioning over speculation. Right now, the agent commerce narrative is in a consolidation phase. The market is waiting for a catalyst: either a high-profile agent shopping error that triggers a regulatory panic, or a breakthrough in AI safety that makes agents fully trusted. Either outcome will redefine the winners.
My bet is on the crypto-native infrastructure. Visa will always be a partner, not a disruptor. The future of agent commerce will look more like a DeFi pool—permissionless, composable, and trust-minimized—than a Visa card swiping in a terminal.
Buy the fear in crypto payment rails. Code the future of autonomous value transfer. Risk is a variable—not a verdict. And right now, the variable favors the side without the middleman.
Watch these signals: - Monthly on-chain volume from x402 and MPP protocols - Consumer trust surveys from Product.ai: if the 14% trust rate hits 30%, the narrative shifts - Visa fiscal Q3 2027 agent commerce transaction count (if they break it out)
Until then, the infrastructure is ahead of demand. I’ve seen this movie before. The battle traders win by positioning early, not by betting on the house.