At Token Relations, we’re always trying to push the envelope with original analysis and unique perspectives. We showcase this through a number of newsletters for clients ranging from Solana to Ripple and in sector specific updates focused on tokenization, DeFi, stablecoins and institutional ethereum. 

But we didn’t have an outlet for pure, topical analyst insights and opinions. Until now. 

Say hello to Strata Research. A new production by our institutional research team that focuses on current, relevant topics with the aim to inform, educate and start conversations. 

Our first edition is focused on agentic payments, a space that everyone from Coinbase to BlackRock is talking about. If there’s a specific topic or theme you want us to cover, simply reply to this email. 

An honest read of the current moment will reveal that vast infrastructure for agentic payments is being built while the volume remains small. The frequent comparison to the early internet is fair but slightly inaccurate, as the web’s infrastructure phase was characterized by uncertainty about which standards would win. Here, growth is characterized more by uncertainty about which flows will scale first and which rails will carry them. 

The flows are differentiating along various lines: stablecoins where the economics favor them; card and bank rails where the trust and dispute infrastructure already exists; and a wallet and policy layer trying to simplify the user experience.

The low-volume data does not say that agentic commerce is failing; instead, it reveals that the infrastructure being built today is mostly for a market that does not yet exist, on the assumption that the trust and model conditions will resolve over the next several years. 

The interesting analytical work, both for crypto-native investors and for traditional finance teams watching the space, will be in tracking which of the four conditions is moving faster than expected, and how that will shape the market when the volume finally arrives.

In December 2025, Bain predicted that by 2030, 15% to 25% of the total U.S. e-commerce market would be handled by AI agents. In dollar terms, that would be anywhere between $300 billion to $500 billion routed by AI. 

That’s an exciting prediction, but it’s a long way away. Today, x402, the most adopted onchain agent payment protocol, clears roughly 180,000 transactions every day, which are altogether worth less than $10,000. It’s not the only agentic payments protocol in use, but the low volume is proof that the industry is far from reaching its potential scale.

Beyond crypto, the picture is similar. Bain itself notes that most U.S. shoppers are not yet comfortable letting AI agents handle a full transaction, and that the future it predicts will require a meaningful step change to be realized. 

For Bain's projection to land, we assert that there are four conditions that need to hold: 

The rest of this analysis works through each of the four conditions, breaking down where progress is being made, and where it isn’t.

Before we dive too deep into where agentic payments are going, let’s quickly go over the basics. If you are already familiar with the different types of agentic payments and how they work, feel free to move on to the later sections.

Agentic payments involve rails and protocols that allow an AI agent to buy something on a user’s behalf, whether that user is a human, a business, or another agent. 

A few use cases of agentic payments include: 

In each of these scenarios, agents select, negotiate, and settle the transaction without requiring a person to approve each step.

However, which rail the agent uses depends on what is being bought. 

For large, discrete payments, like placing an order, booking a reservation, or paying a vendor, agents will likely keep using card networks and bank rails. Users and merchants are used to these avenues, and the chargeback, dispute, and rewards infrastructure built on debit and credit cards is incentive enough for everyone to keep using them. 

But for micropayments, such as buying specific data from behind a paywall, calling an API, or machine-to-machine settlements, stablecoins have a good chance of becoming a preferred avenue. Credit cards are too slow and expensive to handle these transactions, while stablecoins have no trouble keeping up, as they can easily be divided down to fractions of a cent, cheap to send, and can clear in milliseconds, depending on the chain.

On a more technical level, the agentic payments stack can be broken down into roughly seven layers:

These layers do not all face the same end-customer, and the flows running across them split into three economically distinct categories that explain why no single payment protocol or rail has emerged as a winner yet.

Human → Agent → Merchant: A user asks ChatGPT, Gemini or Claude to buy something, and the agent finds the product, negotiates, and settles. The dominant protocols in use here are ACP, AP2 and UCP, and the rails are largely the existing card networks adapted for use by AI agents.

Agent → API/Agent: One agent pays another agent or an API for compute, data, search, or a specialized capability. These are typically sub-cent transactions, and need to clear in milliseconds. This is where stablecoins on fast chains have a major advantage. The unit economics rule out cards and ACH, which is why stablecoins on fast chains carry most of this flow.

Agent → Business Counterparty: This flow is most important for finance teams and covers procurement, payroll and vendor settlement. The rails here are predominantly card issuers, with a newer set of entrants (e.g. Meow, Payman, Catena Labs), wrapping bank accounts in agent-friendly APIs. Stablecoins show up here mostly in cross-border contexts and in vendor settlement for crypto-native counterparties.

Now that we have a firm understanding of the stack, here’s what’s required to grow the agentic payments market. 

The payment-protocol layer currently has more candidates vying for dominance than it can support. Furthermore, the agentic commerce thesis depends on these protocols becoming interoperable enough that an agent can transact across merchants and APIs without requiring bespoke integrations for each. 

That does not necessarily mean we need a single winner. In fact, the most plausible outcome is something like what we have for email, where multiple standards (SMTP, IMAP and MIME) coexist behind a common abstracted layer. Still, the rough number of standards needs to come down, or the integration cost across the long tail of merchants will keep adoption slow.

Merchants and banks will not underwrite AI agent activity at scale until they have a reliable way to identify the agent, who it is representing, what authority it has, and who is liable when something goes wrong. 

ERC-8004, Web Bot Auth, and offerings from Prove, and Experian Agent Trust are all early attempts to build that layer, but none has the kind of cross-vendor support that, say, EMV has for chip-and-PIN for contact and contactless payments with cards and smartphones. Until the trust layer matures, large-ticket consumer and enterprise flows will remain bottlenecked.

Settlement rails must become capable of handling both large and small transactions. The micropayment end has been broadly solved, at least at current volumes; stablecoins on fast L2s and Solana clear sub-cent transactions in milliseconds for fractions of a cent in fees. The large-ticket end has also been solved thanks to efforts from Visa and Mastercard. 

The middle ground, where an agent might want to clear a $20 SaaS subscription or a $200 hardware purchase with the same UX as a $0.05 API call, is where the new wallet and policy layers (Crossmint, Privy, Turnkey, Coinbase Agentic Wallets, Fireblocks) are working to close the gap.

Consumers and finance teams will not hand spending authority to AI agents unless they know it can be trusted. Even setting aside the trust infrastructure, an agent that hallucinates an order line or executes a duplicate transaction once a month cannot be trusted with large amounts of money, regardless of how robust the payment plumbing underneath it is. 

The trajectory of model reliability is therefore paramount to the agentic payments thesis, and it is not one that protocol designers control. Bain's caveat about consumer trust is, in effect, a caveat about model behavior.

The clearest signals of growth in the agentic payments market right now is not user adoption metrics or volume, but infrastructure announcements. Over the past few months, a number of integrations and products have begun to close the gaps in the agentic payments stack. 

Altogether, these announcements have a clear directional bias: nearly all of them deepen the infrastructure for the Agent → API flow, where the rails are stablecoin-native and the unit economics already favor a protocol like x402. (For more details on x402, check out the Talking Tokens episode with its creator Erik Reppel.)

The Human → Agent → Merchant flow has seen progress (e.g., Visa Intelligent Commerce, Mastercard Agent Pay, and Shopify's UCP), but the pace of integration is slower, in part because there are harder problems to solve with identity and trust. 

The Agent → Business flow has the fewest publicly visible integrations, which is consistent with how enterprise software typically rolls out. It also accounts for finance and procurement teams being unlikely to hand over budget authority until the underlying models and policy controls mature.

The machine-to-machine layer is being built first, in part because it has the cleanest fit with existing crypto rails, and partly because the trust problem is easier to solve. An agent paying another agent for a search result is a lower-stakes transaction than an agent buying a $900 smartphone using a credit card. 

Consumer and enterprise volume will arrive later, and most of it will not look like x402 traffic when it does. It will look like card payments and ACH transfers.

This report was authored by Institutional Research Analyst Alexander Beaudry and brought to you by StrataMedia (home to Token Relations, Talking Tokens & More). 

This information is for educational purposes only. It should not be considered financial advice, nor should it be used to make investment decisions. Cryptocurrencies are high risk and you should consult a financial professional before making any financial decisions. Make sure you do your own research.