There is a mystical belief in Silicon Valley that when a major technological shift occurs, past structures wither away. I’ve always thought of these shifts more like a game of jacks. Existing structures are thrown up into the air while everyone gawks at the sky. But the jacks always land, maybe just a bit more scattered than before. Craigslist may have killed newspaper classifieds, but ad budgets migrated to Google and Facebook.

Agentic commerce is an example of one of these shifts.

A credit card transaction today costs merchants roughly 3% or more, including interchange, network fees, and processor margin. For decades, this has been considered the implicit cost of moving money. This is a psyop. Moving money is cheap—ACH costs pennies and wires cost dollars.

Interchange is actually the implicit cost of trust. The issuing bank underwrites credit risk. Fraud protection is guaranteed by the network. The acquirer provides merchant guarantees and chargeback infrastructure. As rails, agentic payment protocols can zero out the cost of moving money, but they don’t zero out the cost of trust in the same way.

x402 is HTTP-native and stablecoin-native, with no PSP in the loop. A site returns a 402 with a price, the client signs a USDC transfer, and it settles onchain near real time. There’s no typical merchant account, acquirer, interchange stack, diverting the existing payment stack. x402’s volume is roughly $1.5M/month.

Agentic Commerce Protocol (ACP), designed by Stripe and OpenAI, standardizes agent product discovery and purchase intent expression, but it delegates payment execution to the PSP. Stripe’s 2.9% survives. Interchange survives. ACP changes who initiates the transaction, not how money settles. 

Machine Payments Protocol (MPP), designed by Tempo and Stripe, can work in either direction: near-free onchain settlement or fiat. Same protocol, radically different economics depending on the exit.

x402 collapses the fee. ACP preserves it. MPP goes either way. 

Taken together, these protocols display fee compression in motion. The incumbents know this; it’s why PayPal, Visa, and Revolut are all launching stablecoin products that may cannibalize their own interchange. When the cost of moving money collapses to zero, the fees must move to other parts of the stack. The question becomes: What new layers of the stack capture those fees? 

‍

1. Agent Identity

The relevance of identity is one of the most hotly debated topics in agentic payments. The common counterargument: instant settlement means no counterparty risk to underwrite, rendering identity irrelevant. For sub-dollar API calls between known services, that’s probably right.

Agent identity isn’t just important for settlement risk. It answers the question: Did this version of this agent have permission to make this payment?  This is an important compliance question that doesn’t go away with instant settlement. Agent authorization is temporal (revoked mid-session), hierarchical (agents delegating to subagents), and non-biometric (so far). Whereas a card payment requires a one-time credential check, an agentic payment requires a claim about current state that must be generated and verified on every transaction. This places identity in the transaction critical path and gives it a durable cost, regardless of how cheap settlement becomes. Visa’s TAP and ERC-8004 are first passes, but the fee here is a per-transaction attestation charge. These fees are small individually, but enormous in aggregate across billions of machine-initiated transactions.

‍

2. Dispute Resolution 

Agents reduce human fraud, but introduce a fundamentally new class of transaction error: systematic error. This is the case where a machine operating in good faith buys the wrong thing or overpays. The existing chargeback framework doesn’t map cleanly where the current four-party card model (issuer, network, acquirer, merchant) assumes a human initiated the transaction. When an agent introduces a fifth participant, that assumption is broken, and liability defaults to the merchant even when the error originated in the agent’s logic. Chargebacks already cost $30B+ per year. Imagine the scale of agent-initiated disputes at machine speed. Whoever builds the arbitration layer for agent-initiated transactions captures a premium toll on every disputed payment.

‍

3. Conversion Optimization 

When agents become the primary shopping interface, the traditional e-commerce funnel inverts. Agents aren’t persuaded by banner ads or visual merchandising. They query data, filter out noise, rank options based on signal, and buy. Agentic discovery and agent conversion optimization are quickly becoming one of the most strategically important dynamics in agentic commerce. Merchants are paying to be agent-readable and agent-optimized. Shopify already has Storefront MCP; Profound is already a $1B company. The businesses best positioned are those natively queryable via API, not legacy retailers retrofitting structured data onto unstructured catalogs. The platform that orchestrates agent-merchant matching across ranking, filtering, and personalized recommendation charges the new merchant discount rate. The addressable pool is the $500B+ that currently flows through Google and Meta ads.

‍

4. Spend Governance 

While compelling, none of the above layers have the structural properties of the next interchange. Identity tells you who an agent is. Dispute resolution happens after the fact. Conversion optimization is an evolution of demand-side marketing. The layer that captures the real toll—the one that sits in the critical path of every transaction regardless of rail, regardless of settlement currency, regardless of whether the agent is buying a $.003 API call or a $50K SaaS contract—is spend governance.

When a CFO hands an agent a $10M annual procurement budget, a wallet with a balance isn’t enough. They need a policy engine that allows them to dictate spending caps by category, preferred suppliers, restricted jurisdictions, human escalation above thresholds, etc. Ramp and Brex do some version of this, allowing you to set merchant-category restrictions on virtual cards. But you can’t give an agent and each of its 52 subagents a scoped wallet/sub-wallet with different spending rules, smart contract escrow, and programmable approval chains on a corporate card. The card abstraction literally can’t express that policy surface, and card rails can’t handle the transaction types that surface will need to authorize. A $0.003 API call or metered compute charge is structurally incompatible with a $0.30 fixed fee per swipe. Moving onchain allows you to do this, and that’s not a fee argument, it’s a capability unlock that traditional rails structurally cannot match. ERC-4337 account abstraction starts to enable this on the wallet side, but the policy layer that defines rules and audits compliance doesn’t exist yet.

Governance wins for three reasons. 

1. It’s rail-agnostic and adoption-gating. Whether an agent pays via x402, ACP, or MPP (whether it settles in stablecoins or fiat), it still needs authorization. That’s a real chokepoint that needs to get solved.
2. It matters now, not just in a far-off agent-to-agent centric future. Today’s agentic payment flows are almost entirely human → agent → API. But even a human-supervised agent spending corporate budget through a harness needs policy enforcement. This is the crucial gap between “agents can pay” and “agents are allowed to pay.”
3. It compounds. Every organization’s spending policies generate data about what good policy design looks like, which improves defaults for the next customer, which drives adoption, which generates more policy data, etc. A classic flywheel.

The obvious question: Why can’t governance also be commoditized? Governance is fundamentally a UX and orchestration problem, rather than a protocol problem. HTTP is an open standard; Google still captures $300B+ in search revenue on top of it. The fee will live at the platform layer because each organization’s control requirements are unique, shaped by its own governance structure, spending budgets, risk tolerance, etc. A protocol can standardize how permissions are expressed; it can’t standardize what permissions an organization needs.

The fee in an agentic future won’t be a flat percentage or a fixed tax on transactions—it’ll be priced on policy complexity, transaction volume, and metered usage. The platform layers of the stack are what matter most: sitting above the rails and charging for the same function interchange has always captured—enforcing the rules around when money is permitted to move.

The jacks have been thrown. Everyone’s watching them in the air, arguing about where they float. But the game is won by whoever reads where they land.

Thank you to Dmitriy Berenzon, Eskender Abebe, Aadharsh Pannirselvam, and Tyler Gehringer for review and feedback on this post. Many thanks to the teams at Merit Systems, Natural Pay, and Rova for conversations that also deeply inspired this post.

‍—

Disclaimer:

This post is for general information purposes only. It does not constitute investment advice or a recommendation or solicitation to buy or sell any investment and should not be used in the evaluation of the merits of making any investment decision. It should not be relied upon for accounting, legal or tax advice or investment recommendations. You should consult your own advisers as to legal, business, tax, and other related matters concerning any investment or legal matters. Certain information contained in here has been obtained from third-party sources, including from portfolio companies of funds managed by Archetype. This post reflects the current opinions of the authors and is not made on behalf of Archetype or its affiliates and does not necessarily reflect the opinions of Archetype, its affiliates or individuals associated with Archetype. The opinions reflected herein are subject to change without being updated.