This week Cuy Sheffield is joined this week by Noah Levine, Partner at a16z; gmoney, crypto investor and longtime NFT collector; and Michael Blau, Head of Product at Royal.

What begins with x402 and agentic payments quickly lands on a sharper point: the first real machine economy may be financial before it is retail. The users most willing to tolerate friction are the ones trying to make money, which makes trading, investing, and usage-based software a more credible early market than the usual “your AI buys groceries” story.

🎙️ Listen to the latest episode of Tokenized here, or📷 watch on YouTube.

Noah Levine’s framing of the “headless merchant economy” was the most useful way to think about the opening x402 discussion. His argument was that many people still hear “agentic commerce” and picture a chatbot buying groceries or sneakers. But the more immediate opportunity may be much less consumer-facing. It may be the next class of developers - including people who have never written code before but are now building with Claude Code or Codex - buying access to tools, services, and APIs on a per-usage basis.

That matters because it shifts the story away from consumer checkout and toward infrastructure. In Noah’s framing, the new merchant may look less like a website with a sales team and annual contract and more like an endpoint that software can call and pay for directly. That is a much more believable early market. The encouraging part, as he noted in the x402 context, is that this is no longer just a long-tail startup thesis. Large companies like Google and AWS are now showing up too.

Michael Blau’s own work on DripStack gave that thesis a more concrete shape. He described first experimenting with an agent-accessible restaurant reservation tool, then moving toward a product that lets writers sell paywalled content directly to agents. The point was not just that AI agents might buy things. It was that they may buy highly specific inputs - data, content, access, and functionality - without the traditional packaging sitting around them.

One theme kept coming back throughout the discussion: the people most willing to push through friction are usually the ones who think there is money on the other side.

Blau made that point directly when he said early demand is likely to come from traders and investors, because they have already shown they will tolerate clunky tools if those tools give them an edge. gmoney came at the same idea from a different angle. He described his own shift from using AI for productivity to asking a more direct question: how do I use these tools to actually make money?

That behavioral shift matters. It suggests the early wallet user may not be the average retail consumer, but the trader, investor, power user, or one-person operator willing to test new tooling if there is a plausible return attached. gmoney’s benchmark was simple: if a $200-a-month Claude plan can help you make more than $200 a month, the logic starts to work.

Another strong point from the episode was Blau’s argument that pay-per-use commerce is not really new. We have seen versions of it before - pay-per-view, $0.99 songs, paid apps. The problem was never that the model was impossible. The problem was that there was too much mental friction for the user. Each small purchase still required a decision.

His point was that AI changes that. If an agent can abstract away the act of deciding whether to buy the next unit of content, data, or software usage, then the old subscription logic starts to weaken. That is what makes pay-per-use interesting again.

The discussion also widened into data markets. Noah’s point was that if compute and data are the key inputs for AI, then creators and specialists may become more economically important than they look today. In that world, agents may pay for differentiated information directly rather than relying only on what is already free on the internet.

Cuy Sheffield framed one of the most practical tensions in the episode. Even as a crypto-native user, he said he would still often rather give an agent a card than ask a mainstream user to manage stablecoins directly. Cards offer familiar controls. They avoid the onboarding burden of setting up wallets and buying stablecoins in the first place.

But he also made the other side of the argument just as clearly. If agentic commerce becomes a high-frequency, pay-as-you-go environment, it makes very little sense for settlement to happen three days later, with weekends and bank holidays still getting in the way. That is where stablecoins start to look much more compelling.

His conclusion - and one of the more institutionally useful ideas in the episode - was the hybrid model: cards on the front end, stablecoins on the back end. That feels like a realistic path. Users may not want a crypto-native experience. Merchants and platforms may still want crypto-native settlement.

That distinction matters. It suggests stablecoins may win less as a visible consumer product and more as infrastructure hidden beneath more familiar interfaces.

The conversation on tokenized assets was stronger than the usual efficiency argument. When Cuy asked what actually drives demand for tokenized stocks versus traditional ones, Noah’s answer was simple: tokenized assets may grow largely as a second-order effect of stablecoin adoption.

His logic was that a growing class of fintechs now gives users access to dollars globally through stablecoins. Once that value is already sitting onchain, the next question is how to invest it. At that point, the easiest investment product to access may be one that sits on the same ledger.

The conversation then pushed that thesis one step further. If every agent ends up with a wallet, tokenized assets may also become easier for software to purchase and manage than traditional brokerage-based assets.

The MoonPay acquisition of Dawn Labs mattered here less as a corporate news item and more as a signal about where AI-native finance tooling is heading. Cuy framed it as part of a broader move toward “vibe trading” and “vibe investing,” where users build a thesis with AI and then ask software to execute it.

Noah argued that this starts to look like the next generation of wealth management: AI helps formulate the strategy, carry out the trades, and update the thesis as new information comes in. But the conversation was careful not to flatten that into a clean success story.

Blau laid out the less serious version too. People already ask ChatGPT what stock to buy, and because the model is often drawing on the same public internet chatter, many users may end up getting pointed toward the same consensus trades. That can become self-fulfilling, but it is not the same thing as serious analysis.

That tension was useful. The shallow version of AI trading is basically automated consensus. The more interesting version depends on better inputs - paid research, specialist writers, and differentiated analysis. That is where Blau’s Dripstack thesis comes back in. If people want more than a model remixing Reddit, the market for financial writing and premium information may become more valuable, not less.

The later part of the conversation zoomed back out. Noah noted that stablecoins now exceed $270 billion in supply and argued that the next phase of crypto adoption is likely to come more from institutions than from retail traders. He also gave a useful checklist of what institutions actually want from onchain systems: predictable fees, stablecoin-denominated payments rather than volatile gas tokens, high throughput, low fees, and privacy.

That is worth paying attention to because it makes the adoption story more concrete. This is less about generic institutional interest now and more about the feature set required to move real flows onchain.

The discussion around Circle’s Arc and non-USD stablecoins pushed on a different but related question: does the market eventually broaden beyond a heavily dollarized stablecoin system? Cuy raised the possibility that Circle’s narrower lineup - effectively USDC and EURC - could leave room for ecosystems built around onchain FX between USDC and other fiat-backed stablecoins.

Blau’s answer was practical. As a builder, he thinks first about liquidity and integrations. If another stablecoin has real liquidity and meaningful support across DeFi and yield venues, he is happy to integrate it. Until then, builders follow the deepest pools. gmoney made a similar point more bluntly: most of these markets, especially stablecoins, tend toward a winner-take-most structure.

That leaves the non-dollar story unresolved, but in a useful way. The strategic case for onchain FX is clearly there. The operational case still depends on whether liquidity actually shows up.

What made this episode useful was that it stayed close to behavior. The conversation was not really about whether AI, stablecoins, or tokenized assets are each individually important. It was about what happens when they start to reinforce each other.

The answer, at least from this discussion, is that the first machine economy may emerge in places where users already tolerate friction in exchange for upside: trading, investing, developer tooling, paid data, and specialized content. If that is right, then stablecoins may matter first as settlement infrastructure, tokenized assets may arrive as the next product layer, and the most important merchants in this new economy may not look much like merchants at all.

Thanks so much for reading the Tokenized Newsletter!

Please share this edition or share it with your colleagues if you enjoyed it!

Disclaimers